SDValue Res = DAG.getTargetConstant(Flag, MVT::i32);
Ops.push_back(Res);
+ unsigned SP = TLI.getStackPointerRegisterToSaveRestore();
for (unsigned Value = 0, Reg = 0, e = ValueVTs.size(); Value != e; ++Value) {
unsigned NumRegs = TLI.getNumRegisters(*DAG.getContext(), ValueVTs[Value]);
MVT RegisterVT = RegVTs[Value];
for (unsigned i = 0; i != NumRegs; ++i) {
assert(Reg < Regs.size() && "Mismatch in # registers expected");
- Ops.push_back(DAG.getRegister(Regs[Reg++], RegisterVT));
+ unsigned TheReg = Regs[Reg++];
+ Ops.push_back(DAG.getRegister(TheReg, RegisterVT));
+
+ // Notice if we clobbered the stack pointer. Yes, inline asm can do this.
+ if (TheReg == SP && Code == InlineAsm::Kind_Clobber) {
+ MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
+ MFI->setHasInlineAsmWithSPAdjust(true);
+ }
}
}
}
AA = &aa;
GFI = gfi;
LibInfo = li;
- TD = DAG.getTarget().getDataLayout();
+ DL = DAG.getTarget().getDataLayout();
Context = DAG.getContext();
LPadToCallSiteMap.clear();
}
PendingExports.clear();
CurInst = NULL;
HasTailCall = false;
+ SDNodeOrder = LowestSDNodeOrder;
}
/// clearDanglingDebugInfo - Clear the dangling debug information
MachineBasicBlock *TBB,
MachineBasicBlock *FBB,
MachineBasicBlock *CurBB,
- MachineBasicBlock *SwitchBB) {
+ MachineBasicBlock *SwitchBB,
+ uint32_t TWeight,
+ uint32_t FWeight) {
const BasicBlock *BB = CurBB->getBasicBlock();
// If the leaf of the tree is a comparison, merge the condition into
}
CaseBlock CB(Condition, BOp->getOperand(0),
- BOp->getOperand(1), NULL, TBB, FBB, CurBB);
+ BOp->getOperand(1), NULL, TBB, FBB, CurBB, TWeight, FWeight);
SwitchCases.push_back(CB);
return;
}
// Create a CaseBlock record representing this branch.
CaseBlock CB(ISD::SETEQ, Cond, ConstantInt::getTrue(*DAG.getContext()),
- NULL, TBB, FBB, CurBB);
+ NULL, TBB, FBB, CurBB, TWeight, FWeight);
SwitchCases.push_back(CB);
}
+/// Scale down both weights to fit into uint32_t.
+static void ScaleWeights(uint64_t &NewTrue, uint64_t &NewFalse) {
+ uint64_t NewMax = (NewTrue > NewFalse) ? NewTrue : NewFalse;
+ uint32_t Scale = (NewMax / UINT32_MAX) + 1;
+ NewTrue = NewTrue / Scale;
+ NewFalse = NewFalse / Scale;
+}
+
/// FindMergedConditions - If Cond is an expression like
void SelectionDAGBuilder::FindMergedConditions(const Value *Cond,
MachineBasicBlock *TBB,
MachineBasicBlock *FBB,
MachineBasicBlock *CurBB,
MachineBasicBlock *SwitchBB,
- unsigned Opc) {
+ unsigned Opc, uint32_t TWeight,
+ uint32_t FWeight) {
// If this node is not part of the or/and tree, emit it as a branch.
const Instruction *BOp = dyn_cast<Instruction>(Cond);
if (!BOp || !(isa<BinaryOperator>(BOp) || isa<CmpInst>(BOp)) ||
BOp->getParent() != CurBB->getBasicBlock() ||
!InBlock(BOp->getOperand(0), CurBB->getBasicBlock()) ||
!InBlock(BOp->getOperand(1), CurBB->getBasicBlock())) {
- EmitBranchForMergedCondition(Cond, TBB, FBB, CurBB, SwitchBB);
+ EmitBranchForMergedCondition(Cond, TBB, FBB, CurBB, SwitchBB,
+ TWeight, FWeight);
return;
}
if (Opc == Instruction::Or) {
// Codegen X | Y as:
+ // BB1:
// jmp_if_X TBB
// jmp TmpBB
// TmpBB:
// jmp FBB
//
+ // We have flexibility in setting Prob for BB1 and Prob for TmpBB.
+ // The requirement is that
+ // TrueProb for BB1 + (FalseProb for BB1 * TrueProb for TmpBB)
+ // = TrueProb for orignal BB.
+ // Assuming the orignal weights are A and B, one choice is to set BB1's
+ // weights to A and A+2B, and set TmpBB's weights to A and 2B. This choice
+ // assumes that
+ // TrueProb for BB1 == FalseProb for BB1 * TrueProb for TmpBB.
+ // Another choice is to assume TrueProb for BB1 equals to TrueProb for
+ // TmpBB, but the math is more complicated.
+
+ uint64_t NewTrueWeight = TWeight;
+ uint64_t NewFalseWeight = (uint64_t)TWeight + 2 * (uint64_t)FWeight;
+ ScaleWeights(NewTrueWeight, NewFalseWeight);
// Emit the LHS condition.
- FindMergedConditions(BOp->getOperand(0), TBB, TmpBB, CurBB, SwitchBB, Opc);
+ FindMergedConditions(BOp->getOperand(0), TBB, TmpBB, CurBB, SwitchBB, Opc,
+ NewTrueWeight, NewFalseWeight);
+ NewTrueWeight = TWeight;
+ NewFalseWeight = 2 * (uint64_t)FWeight;
+ ScaleWeights(NewTrueWeight, NewFalseWeight);
// Emit the RHS condition into TmpBB.
- FindMergedConditions(BOp->getOperand(1), TBB, FBB, TmpBB, SwitchBB, Opc);
+ FindMergedConditions(BOp->getOperand(1), TBB, FBB, TmpBB, SwitchBB, Opc,
+ NewTrueWeight, NewFalseWeight);
} else {
assert(Opc == Instruction::And && "Unknown merge op!");
// Codegen X & Y as:
+ // BB1:
// jmp_if_X TmpBB
// jmp FBB
// TmpBB:
//
// This requires creation of TmpBB after CurBB.
+ // We have flexibility in setting Prob for BB1 and Prob for TmpBB.
+ // The requirement is that
+ // FalseProb for BB1 + (TrueProb for BB1 * FalseProb for TmpBB)
+ // = FalseProb for orignal BB.
+ // Assuming the orignal weights are A and B, one choice is to set BB1's
+ // weights to 2A+B and B, and set TmpBB's weights to 2A and B. This choice
+ // assumes that
+ // FalseProb for BB1 == TrueProb for BB1 * FalseProb for TmpBB.
+
+ uint64_t NewTrueWeight = 2 * (uint64_t)TWeight + (uint64_t)FWeight;
+ uint64_t NewFalseWeight = FWeight;
+ ScaleWeights(NewTrueWeight, NewFalseWeight);
// Emit the LHS condition.
- FindMergedConditions(BOp->getOperand(0), TmpBB, FBB, CurBB, SwitchBB, Opc);
+ FindMergedConditions(BOp->getOperand(0), TmpBB, FBB, CurBB, SwitchBB, Opc,
+ NewTrueWeight, NewFalseWeight);
+ NewTrueWeight = 2 * (uint64_t)TWeight;
+ NewFalseWeight = FWeight;
+ ScaleWeights(NewTrueWeight, NewFalseWeight);
// Emit the RHS condition into TmpBB.
- FindMergedConditions(BOp->getOperand(1), TBB, FBB, TmpBB, SwitchBB, Opc);
+ FindMergedConditions(BOp->getOperand(1), TBB, FBB, TmpBB, SwitchBB, Opc,
+ NewTrueWeight, NewFalseWeight);
}
}
(BOp->getOpcode() == Instruction::And ||
BOp->getOpcode() == Instruction::Or)) {
FindMergedConditions(BOp, Succ0MBB, Succ1MBB, BrMBB, BrMBB,
- BOp->getOpcode());
+ BOp->getOpcode(), getEdgeWeight(BrMBB, Succ0MBB),
+ getEdgeWeight(BrMBB, Succ1MBB));
// If the compares in later blocks need to use values not currently
// exported from this block, export them now. This block should always
// be the first entry.
volatile double RDensity =
(double)RSize.roundToDouble() /
(Last - RBegin + 1ULL).roundToDouble();
- double Metric = Range.logBase2()*(LDensity+RDensity);
+ volatile double Metric = Range.logBase2()*(LDensity+RDensity);
// Should always split in some non-trivial place
DEBUG(dbgs() <<"=>Step\n"
<< "LEnd: " << LEnd << ", RBegin: " << RBegin << '\n'
if (DestVT != N.getValueType())
setValue(&I, DAG.getNode(ISD::BITCAST, getCurSDLoc(),
DestVT, N)); // convert types.
+ // Check if the original LLVM IR Operand was a ConstantInt, because getValue()
+ // might fold any kind of constant expression to an integer constant and that
+ // is not what we are looking for. Only regcognize a bitcast of a genuine
+ // constant integer as an opaque constant.
+ else if(ConstantInt *C = dyn_cast<ConstantInt>(I.getOperand(0)))
+ setValue(&I, DAG.getConstant(C->getValue(), DestVT, /*isTarget=*/false,
+ /*isOpaque*/true));
else
setValue(&I, N); // noop cast.
}
unsigned Field = cast<Constant>(Idx)->getUniqueInteger().getZExtValue();
if (Field) {
// N = N + Offset
- uint64_t Offset = TD->getStructLayout(StTy)->getElementOffset(Field);
+ uint64_t Offset = DL->getStructLayout(StTy)->getElementOffset(Field);
N = DAG.getNode(ISD::ADD, getCurSDLoc(), N.getValueType(), N,
DAG.getConstant(Offset, N.getValueType()));
}
if (const ConstantInt *CI = dyn_cast<ConstantInt>(Idx)) {
if (CI->isZero()) continue;
uint64_t Offs =
- TD->getTypeAllocSize(Ty)*cast<ConstantInt>(CI)->getSExtValue();
+ DL->getTypeAllocSize(Ty)*cast<ConstantInt>(CI)->getSExtValue();
SDValue OffsVal;
EVT PTy = TLI->getPointerTy(AS);
unsigned PtrBits = PTy.getSizeInBits();
// N = N + Idx * ElementSize;
APInt ElementSize = APInt(TLI->getPointerSizeInBits(AS),
- TD->getTypeAllocSize(Ty));
+ DL->getTypeAllocSize(Ty));
SDValue IdxN = getValue(Idx);
// If the index is smaller or larger than intptr_t, truncate or extend
// Inform the Frame Information that we have just allocated a variable-sized
// object.
- FuncInfo.MF->getFrameInfo()->CreateVariableSizedObject(Align ? Align : 1);
+ FuncInfo.MF->getFrameInfo()->CreateVariableSizedObject(Align ? Align : 1, &I);
}
void SelectionDAGBuilder::visitLoad(const LoadInst &I) {
SDValue Root;
bool ConstantMemory = false;
- if (I.isVolatile() || NumValues > MaxParallelChains)
+ if (isVolatile || NumValues > MaxParallelChains)
// Serialize volatile loads with other side effects.
Root = getRoot();
else if (AA->pointsToConstantMemory(
Root = DAG.getRoot();
}
+ const TargetLowering *TLI = TM.getTargetLowering();
+ if (isVolatile)
+ Root = TLI->prepareVolatileOrAtomicLoad(Root, getCurSDLoc(), DAG);
+
SmallVector<SDValue, 4> Values(NumValues);
SmallVector<SDValue, 4> Chains(std::min(unsigned(MaxParallelChains),
NumValues));
if (I.getAlignment() < VT.getSizeInBits() / 8)
report_fatal_error("Cannot generate unaligned atomic load");
+ InChain = TLI->prepareVolatileOrAtomicLoad(InChain, dl, DAG);
SDValue L =
DAG.getAtomic(ISD::ATOMIC_LOAD, dl, VT, VT, InChain,
getValue(I.getPointerOperand()),
static SDValue expandPow(SDLoc dl, SDValue LHS, SDValue RHS,
SelectionDAG &DAG, const TargetLowering &TLI) {
bool IsExp10 = false;
- if (LHS.getValueType() == MVT::f32 && LHS.getValueType() == MVT::f32 &&
+ if (LHS.getValueType() == MVT::f32 && RHS.getValueType() == MVT::f32 &&
LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
if (ConstantFPSDNode *LHSC = dyn_cast<ConstantFPSDNode>(LHS)) {
APFloat Ten(10.0f);
return 0;
SmallVector<Value *, 4> Allocas;
- GetUnderlyingObjects(I.getArgOperand(1), Allocas, TD);
+ GetUnderlyingObjects(I.getArgOperand(1), Allocas, DL);
for (SmallVectorImpl<Value*>::iterator Object = Allocas.begin(),
E = Allocas.end(); Object != E; ++Object) {
int DemoteStackIdx = -100;
if (!CanLowerReturn) {
+ assert(!CS.hasInAllocaArgument() &&
+ "sret demotion is incompatible with inalloca");
uint64_t TySize = TLI->getDataLayout()->getTypeAllocSize(
FTy->getReturnType());
unsigned Align = TLI->getDataLayout()->getPrefTypeAlignment(
if (const Constant *LoadCst =
ConstantFoldLoadFromConstPtr(const_cast<Constant *>(LoadInput),
- Builder.TD))
+ Builder.DL))
return Builder.getValue(LoadCst);
}
// bloat the code.
const TargetLowering *TLI = TM.getTargetLowering();
if (ActuallyDoIt && CSize->getZExtValue() > 4) {
+ unsigned DstAS = LHS->getType()->getPointerAddressSpace();
+ unsigned SrcAS = RHS->getType()->getPointerAddressSpace();
// TODO: Handle 5 byte compare as 4-byte + 1 byte.
// TODO: Handle 8 byte compare on x86-32 as two 32-bit loads.
- if (!TLI->isTypeLegal(LoadVT) ||!TLI->allowsUnalignedMemoryAccesses(LoadVT))
+ if (!TLI->isTypeLegal(LoadVT) ||
+ !TLI->allowsUnalignedMemoryAccesses(LoadVT, SrcAS) ||
+ !TLI->allowsUnalignedMemoryAccesses(LoadVT, DstAS))
ActuallyDoIt = false;
}
/// MVT::Other.
EVT getCallOperandValEVT(LLVMContext &Context,
const TargetLowering &TLI,
- const DataLayout *TD) const {
+ const DataLayout *DL) const {
if (CallOperandVal == 0) return MVT::Other;
if (isa<BasicBlock>(CallOperandVal))
// If OpTy is not a single value, it may be a struct/union that we
// can tile with integers.
if (!OpTy->isSingleValueType() && OpTy->isSized()) {
- unsigned BitSize = TD->getTypeSizeInBits(OpTy);
+ unsigned BitSize = DL->getTypeSizeInBits(OpTy);
switch (BitSize) {
default: break;
case 1:
OpInfo.CallOperand = getValue(OpInfo.CallOperandVal);
}
- OpVT = OpInfo.getCallOperandValEVT(*DAG.getContext(), *TLI, TD).
+ OpVT = OpInfo.getCallOperandValEVT(*DAG.getContext(), *TLI, DL).
getSimpleVT();
}
void SelectionDAGBuilder::visitVAArg(const VAArgInst &I) {
const TargetLowering *TLI = TM.getTargetLowering();
- const DataLayout &TD = *TLI->getDataLayout();
+ const DataLayout &DL = *TLI->getDataLayout();
SDValue V = DAG.getVAArg(TLI->getValueType(I.getType()), getCurSDLoc(),
getRoot(), getValue(I.getOperand(0)),
DAG.getSrcValue(I.getOperand(0)),
- TD.getABITypeAlignment(I.getType()));
+ DL.getABITypeAlignment(I.getType()));
setValue(&I, V);
DAG.setRoot(V.getValue(1));
}
assert(CI.getType()->isVoidTy() && "Stackmap cannot return a value.");
- SDValue Callee = getValue(CI.getCalledValue());
+ SDValue Chain, InFlag, Callee, NullPtr;
+ SmallVector<SDValue, 32> Ops;
- // Lower into a call sequence with no args and no return value.
- std::pair<SDValue, SDValue> Result = LowerCallOperands(CI, 0, 0, Callee);
- // Set the root to the target-lowered call chain.
- SDValue Chain = Result.second;
- DAG.setRoot(Chain);
+ SDLoc DL = getCurSDLoc();
+ Callee = getValue(CI.getCalledValue());
+ NullPtr = DAG.getIntPtrConstant(0, true);
- /// Get a call instruction from the call sequence chain.
- /// Tail calls are not allowed.
- SDNode *CallEnd = Chain.getNode();
- assert(CallEnd->getOpcode() == ISD::CALLSEQ_END &&
- "Expected a callseq node.");
- SDNode *Call = CallEnd->getOperand(0).getNode();
- bool hasGlue = Call->getGluedNode();
+ // The stackmap intrinsic only records the live variables (the arguemnts
+ // passed to it) and emits NOPS (if requested). Unlike the patchpoint
+ // intrinsic, this won't be lowered to a function call. This means we don't
+ // have to worry about calling conventions and target specific lowering code.
+ // Instead we perform the call lowering right here.
+ //
+ // chain, flag = CALLSEQ_START(chain, 0)
+ // chain, flag = STACKMAP(id, nbytes, ..., chain, flag)
+ // chain, flag = CALLSEQ_END(chain, 0, 0, flag)
+ //
+ Chain = DAG.getCALLSEQ_START(getRoot(), NullPtr, DL);
+ InFlag = Chain.getValue(1);
- // Replace the target specific call node with the stackmap intrinsic.
- SmallVector<SDValue, 8> Ops;
+ // Add the <id> and <numBytes> constants.
+ SDValue IDVal = getValue(CI.getOperand(PatchPointOpers::IDPos));
+ Ops.push_back(DAG.getTargetConstant(
+ cast<ConstantSDNode>(IDVal)->getZExtValue(), MVT::i64));
+ SDValue NBytesVal = getValue(CI.getOperand(PatchPointOpers::NBytesPos));
+ Ops.push_back(DAG.getTargetConstant(
+ cast<ConstantSDNode>(NBytesVal)->getZExtValue(), MVT::i32));
- // Add the <id> and <numShadowBytes> constants.
- for (unsigned i = 0; i < 2; ++i) {
- SDValue tmp = getValue(CI.getOperand(i));
- Ops.push_back(DAG.getTargetConstant(
- cast<ConstantSDNode>(tmp)->getZExtValue(), MVT::i32));
- }
// Push live variables for the stack map.
addStackMapLiveVars(CI, 2, Ops, *this);
- // Push the chain (this is originally the first operand of the call, but
- // becomes now the last or second to last operand).
- Ops.push_back(*(Call->op_begin()));
+ // We are not pushing any register mask info here on the operands list,
+ // because the stackmap doesn't clobber anything.
- // Push the glue flag (last operand).
- if (hasGlue)
- Ops.push_back(*(Call->op_end()-1));
+ // Push the chain and the glue flag.
+ Ops.push_back(Chain);
+ Ops.push_back(InFlag);
+ // Create the STACKMAP node.
SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
+ SDNode *SM = DAG.getMachineNode(TargetOpcode::STACKMAP, DL, NodeTys, Ops);
+ Chain = SDValue(SM, 0);
+ InFlag = Chain.getValue(1);
- // Replace the target specific call node with a STACKMAP node.
- MachineSDNode *MN = DAG.getMachineNode(TargetOpcode::STACKMAP, getCurSDLoc(),
- NodeTys, Ops);
+ Chain = DAG.getCALLSEQ_END(Chain, NullPtr, NullPtr, InFlag, DL);
- // StackMap generates no value, so nothing goes in the NodeMap.
+ // Stackmaps don't generate values, so nothing goes into the NodeMap.
- // Fixup the consumers of the intrinsic. The chain and glue may be used in the
- // call sequence.
- DAG.ReplaceAllUsesWith(Call, MN);
+ // Set the root to the target-lowered call chain.
+ DAG.setRoot(Chain);
- DAG.DeleteNode(Call);
+ // Inform the Frame Information that we have a stackmap in this function.
+ FuncInfo.MF->getFrameInfo()->setHasStackMap();
}
/// \brief Lower llvm.experimental.patchpoint directly to its target opcode.
void SelectionDAGBuilder::visitPatchpoint(const CallInst &CI) {
- // void|i64 @llvm.experimental.patchpoint.void|i64(i32 <id>,
+ // void|i64 @llvm.experimental.patchpoint.void|i64(i64 <id>,
// i32 <numBytes>,
// i8* <target>,
// i32 <numArgs>,
// Add the <id> and <numBytes> constants.
SDValue IDVal = getValue(CI.getOperand(PatchPointOpers::IDPos));
Ops.push_back(DAG.getTargetConstant(
- cast<ConstantSDNode>(IDVal)->getZExtValue(), MVT::i32));
+ cast<ConstantSDNode>(IDVal)->getZExtValue(), MVT::i64));
SDValue NBytesVal = getValue(CI.getOperand(PatchPointOpers::NBytesPos));
Ops.push_back(DAG.getTargetConstant(
cast<ConstantSDNode>(NBytesVal)->getZExtValue(), MVT::i32));
} else
DAG.ReplaceAllUsesWith(Call, MN);
DAG.DeleteNode(Call);
+
+ // Inform the Frame Information that we have a patchpoint in this function.
+ FuncInfo.MF->getFrameInfo()->setHasPatchPoint();
}
/// TargetLowering::LowerCallTo - This is the default LowerCallTo
Flags.setInReg();
if (Args[i].isSRet)
Flags.setSRet();
- if (Args[i].isByVal) {
+ if (Args[i].isByVal)
+ Flags.setByVal();
+ if (Args[i].isInAlloca) {
+ Flags.setInAlloca();
+ // Set the byval flag for CCAssignFn callbacks that don't know about
+ // inalloca. This way we can know how many bytes we should've allocated
+ // and how many bytes a callee cleanup function will pop. If we port
+ // inalloca to more targets, we'll have to add custom inalloca handling
+ // in the various CC lowering callbacks.
Flags.setByVal();
+ }
+ if (Args[i].isByVal || Args[i].isInAlloca) {
PointerType *Ty = cast<PointerType>(Args[i].Ty);
Type *ElementTy = Ty->getElementType();
Flags.setByValSize(getDataLayout()->getTypeAllocSize(ElementTy));
SelectionDAG &DAG = SDB->DAG;
SDLoc dl = SDB->getCurSDLoc();
const TargetLowering *TLI = getTargetLowering();
- const DataLayout *TD = TLI->getDataLayout();
+ const DataLayout *DL = TLI->getDataLayout();
SmallVector<ISD::InputArg, 16> Ins;
if (!FuncInfo->CanLowerReturn) {
Type *ArgTy = VT.getTypeForEVT(*DAG.getContext());
ISD::ArgFlagsTy Flags;
unsigned OriginalAlignment =
- TD->getABITypeAlignment(ArgTy);
+ DL->getABITypeAlignment(ArgTy);
if (F.getAttributes().hasAttribute(Idx, Attribute::ZExt))
Flags.setZExt();
Flags.setInReg();
if (F.getAttributes().hasAttribute(Idx, Attribute::StructRet))
Flags.setSRet();
- if (F.getAttributes().hasAttribute(Idx, Attribute::ByVal)) {
+ if (F.getAttributes().hasAttribute(Idx, Attribute::ByVal))
Flags.setByVal();
+ if (F.getAttributes().hasAttribute(Idx, Attribute::InAlloca)) {
+ Flags.setInAlloca();
+ // Set the byval flag for CCAssignFn callbacks that don't know about
+ // inalloca. This way we can know how many bytes we should've allocated
+ // and how many bytes a callee cleanup function will pop. If we port
+ // inalloca to more targets, we'll have to add custom inalloca handling
+ // in the various CC lowering callbacks.
+ Flags.setByVal();
+ }
+ if (Flags.isByVal() || Flags.isInAlloca()) {
PointerType *Ty = cast<PointerType>(I->getType());
Type *ElementTy = Ty->getElementType();
- Flags.setByValSize(TD->getTypeAllocSize(ElementTy));
+ Flags.setByValSize(DL->getTypeAllocSize(ElementTy));
// For ByVal, alignment should be passed from FE. BE will guess if
// this info is not there but there are cases it cannot get right.
unsigned FrameAlign;
projects / oota-llvm.git / blobdiff