cl::init(0));
static cl::opt<bool>
-EnableFMFInDAG("enable-fmf-dag", cl::init(false), cl::Hidden,
+EnableFMFInDAG("enable-fmf-dag", cl::init(true), cl::Hidden,
cl::desc("Enable fast-math-flags for DAG nodes"));
// Limit the width of DAG chains. This is important in general to prevent
}
void SelectionDAGBuilder::visitCatchPad(const CatchPadInst &I) {
- // Update machine-CFG edges.
- MachineBasicBlock *PadMBB = FuncInfo.MBB;
- MachineBasicBlock *CatchingMBB = FuncInfo.MBBMap[I.getNormalDest()];
- MachineBasicBlock *UnwindMBB = FuncInfo.MBBMap[I.getUnwindDest()];
- PadMBB->addSuccessor(CatchingMBB);
- PadMBB->addSuccessor(UnwindMBB);
-
- CatchingMBB->setIsEHFuncletEntry();
- MachineModuleInfo &MMI = DAG.getMachineFunction().getMMI();
- MMI.setHasEHFunclets(true);
+ llvm_unreachable("should never codegen catchpads");
}
void SelectionDAGBuilder::visitCatchRet(const CatchReturnInst &I) {
// Update machine-CFG edge.
- MachineBasicBlock *PadMBB = FuncInfo.MBB;
MachineBasicBlock *TargetMBB = FuncInfo.MBBMap[I.getSuccessor()];
- PadMBB->addSuccessor(TargetMBB);
+ FuncInfo.MBB->addSuccessor(TargetMBB);
// Create the terminator node.
SDValue Ret = DAG.getNode(ISD::CATCHRET, getCurSDLoc(), MVT::Other,
}
void SelectionDAGBuilder::visitCatchEndPad(const CatchEndPadInst &I) {
- // If this unwinds to caller, we don't need a DAG node hanging around.
- if (!I.hasUnwindDest())
- return;
-
- // Update machine-CFG edge.
- MachineBasicBlock *PadMBB = FuncInfo.MBB;
- MachineBasicBlock *UnwindMBB = FuncInfo.MBBMap[I.getUnwindDest()];
- PadMBB->addSuccessor(UnwindMBB);
+ llvm_unreachable("should never codegen catchendpads");
}
void SelectionDAGBuilder::visitCleanupPad(const CleanupPadInst &CPI) {
- MachineModuleInfo &MMI = DAG.getMachineFunction().getMMI();
- MMI.setHasEHFunclets(true);
- report_fatal_error("visitCleanupPad not yet implemented!");
+ // Don't emit any special code for the cleanuppad instruction. It just marks
+ // the start of a funclet.
+ FuncInfo.MBB->setIsEHFuncletEntry();
+ FuncInfo.MBB->setIsCleanupFuncletEntry();
+}
+
+/// When an invoke or a cleanupret unwinds to the next EH pad, there are
+/// many places it could ultimately go. In the IR, we have a single unwind
+/// destination, but in the machine CFG, we enumerate all the possible blocks.
+/// This function skips over imaginary basic blocks that hold catchpad,
+/// terminatepad, or catchendpad instructions, and finds all the "real" machine
+/// basic block destinations.
+static void
+findUnwindDestinations(FunctionLoweringInfo &FuncInfo,
+ const BasicBlock *EHPadBB,
+ SmallVectorImpl<MachineBasicBlock *> &UnwindDests) {
+ bool IsMSVCCXX = classifyEHPersonality(FuncInfo.Fn->getPersonalityFn()) ==
+ EHPersonality::MSVC_CXX;
+ while (EHPadBB) {
+ const Instruction *Pad = EHPadBB->getFirstNonPHI();
+ if (isa<LandingPadInst>(Pad)) {
+ // Stop on landingpads. They are not funclets.
+ UnwindDests.push_back(FuncInfo.MBBMap[EHPadBB]);
+ break;
+ } else if (isa<CleanupPadInst>(Pad) || isa<LandingPadInst>(Pad)) {
+ // Stop on cleanup pads. Cleanups are always funclet entries for all known
+ // personalities.
+ UnwindDests.push_back(FuncInfo.MBBMap[EHPadBB]);
+ UnwindDests.back()->setIsEHFuncletEntry();
+ break;
+ } else if (const auto *CPI = dyn_cast<CatchPadInst>(Pad)) {
+ // Add the catchpad handler to the possible destinations.
+ UnwindDests.push_back(FuncInfo.MBBMap[CPI->getNormalDest()]);
+ // In MSVC C++, catchblocks are funclets and need prologues.
+ if (IsMSVCCXX)
+ UnwindDests.back()->setIsEHFuncletEntry();
+ EHPadBB = CPI->getUnwindDest();
+ } else if (const auto *CEPI = dyn_cast<CatchEndPadInst>(Pad)) {
+ EHPadBB = CEPI->getUnwindDest();
+ } else if (const auto *CEPI = dyn_cast<CleanupEndPadInst>(Pad)) {
+ EHPadBB = CEPI->getUnwindDest();
+ }
+ }
}
void SelectionDAGBuilder::visitCleanupRet(const CleanupReturnInst &I) {
- report_fatal_error("visitCleanupRet not yet implemented!");
+ // Update successor info.
+ // FIXME: The weights for catchpads will be wrong.
+ SmallVector<MachineBasicBlock *, 1> UnwindDests;
+ findUnwindDestinations(FuncInfo, I.getUnwindDest(), UnwindDests);
+ for (MachineBasicBlock *UnwindDest : UnwindDests) {
+ UnwindDest->setIsEHPad();
+ addSuccessorWithWeight(FuncInfo.MBB, UnwindDest);
+ }
+
+ // Create the terminator node.
+ SDValue Ret =
+ DAG.getNode(ISD::CLEANUPRET, getCurSDLoc(), MVT::Other, getControlRoot());
+ DAG.setRoot(Ret);
}
void SelectionDAGBuilder::visitCleanupEndPad(const CleanupEndPadInst &I) {
void SelectionDAGBuilder::visitInvoke(const InvokeInst &I) {
MachineBasicBlock *InvokeMBB = FuncInfo.MBB;
- // Retrieve successors.
+ // Retrieve successors. Look through artificial IR level blocks like catchpads
+ // and catchendpads for successors.
MachineBasicBlock *Return = FuncInfo.MBBMap[I.getSuccessor(0)];
- MachineBasicBlock *LandingPad = FuncInfo.MBBMap[I.getSuccessor(1)];
+ const BasicBlock *EHPadBB = I.getSuccessor(1);
const Value *Callee(I.getCalledValue());
const Function *Fn = dyn_cast<Function>(Callee);
break;
case Intrinsic::experimental_patchpoint_void:
case Intrinsic::experimental_patchpoint_i64:
- visitPatchpoint(&I, LandingPad);
+ visitPatchpoint(&I, EHPadBB);
break;
case Intrinsic::experimental_gc_statepoint:
- LowerStatepoint(ImmutableStatepoint(&I), LandingPad);
+ LowerStatepoint(ImmutableStatepoint(&I), EHPadBB);
break;
}
} else
- LowerCallTo(&I, getValue(Callee), false, LandingPad);
+ LowerCallTo(&I, getValue(Callee), false, EHPadBB);
// If the value of the invoke is used outside of its defining block, make it
// available as a virtual register.
CopyToExportRegsIfNeeded(&I);
}
- // Update successor info
+ SmallVector<MachineBasicBlock *, 1> UnwindDests;
+ findUnwindDestinations(FuncInfo, EHPadBB, UnwindDests);
+
+ // Update successor info.
+ // FIXME: The weights for catchpads will be wrong.
addSuccessorWithWeight(InvokeMBB, Return);
- addSuccessorWithWeight(InvokeMBB, LandingPad);
+ for (MachineBasicBlock *UnwindDest : UnwindDests) {
+ UnwindDest->setIsEHPad();
+ addSuccessorWithWeight(InvokeMBB, UnwindDest);
+ }
// Drop into normal successor.
DAG.setRoot(DAG.getNode(ISD::BR, getCurSDLoc(),
void SelectionDAGBuilder::visitUnreachable(const UnreachableInst &I) {
if (DAG.getTarget().Options.TrapUnreachable)
- DAG.setRoot(DAG.getNode(ISD::TRAP, getCurSDLoc(), MVT::Other, DAG.getRoot()));
+ DAG.setRoot(
+ DAG.getNode(ISD::TRAP, getCurSDLoc(), MVT::Other, DAG.getRoot()));
}
void SelectionDAGBuilder::visitFSub(const User &I) {
SDValue Op1 = getValue(I.getOperand(0));
SDValue Op2 = getValue(I.getOperand(1));
ISD::CondCode Condition = getFCmpCondCode(predicate);
+
+ // FIXME: Fcmp instructions have fast-math-flags in IR, so we should use them.
+ // FIXME: We should propagate the fast-math-flags to the DAG node itself for
+ // further optimization, but currently FMF is only applicable to binary nodes.
if (TM.Options.NoNaNsFPMath)
Condition = getFCmpCodeWithoutNaN(Condition);
EVT DestVT = DAG.getTargetLoweringInfo().getValueType(DAG.getDataLayout(),
static SDValue getLimitedPrecisionExp2(SDValue t0, SDLoc dl,
SelectionDAG &DAG) {
+ // TODO: What fast-math-flags should be set on the floating-point nodes?
+
// IntegerPartOfX = ((int32_t)(t0);
SDValue IntegerPartOfX = DAG.getNode(ISD::FP_TO_SINT, dl, MVT::i32, t0);
//
// #define LOG2OFe 1.4426950f
// t0 = Op * LOG2OFe
+
+ // TODO: What fast-math-flags should be set here?
SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, Op,
getF32Constant(DAG, 0x3fb8aa3b, dl));
return getLimitedPrecisionExp2(t0, dl, DAG);
/// limited-precision mode.
static SDValue expandLog(SDLoc dl, SDValue Op, SelectionDAG &DAG,
const TargetLowering &TLI) {
+
+ // TODO: What fast-math-flags should be set on the floating-point nodes?
+
if (Op.getValueType() == MVT::f32 &&
LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
SDValue Op1 = DAG.getNode(ISD::BITCAST, dl, MVT::i32, Op);
/// limited-precision mode.
static SDValue expandLog2(SDLoc dl, SDValue Op, SelectionDAG &DAG,
const TargetLowering &TLI) {
+
+ // TODO: What fast-math-flags should be set on the floating-point nodes?
+
if (Op.getValueType() == MVT::f32 &&
LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
SDValue Op1 = DAG.getNode(ISD::BITCAST, dl, MVT::i32, Op);
/// limited-precision mode.
static SDValue expandLog10(SDLoc dl, SDValue Op, SelectionDAG &DAG,
const TargetLowering &TLI) {
+
+ // TODO: What fast-math-flags should be set on the floating-point nodes?
+
if (Op.getValueType() == MVT::f32 &&
LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
SDValue Op1 = DAG.getNode(ISD::BITCAST, dl, MVT::i32, Op);
}
}
+ // TODO: What fast-math-flags should be set on the FMUL node?
if (IsExp10) {
// Put the exponent in the right bit position for later addition to the
// final result:
// the benefit of being both really simple and much better than a libcall.
SDValue Res; // Logically starts equal to 1.0
SDValue CurSquare = LHS;
+ // TODO: Intrinsics should have fast-math-flags that propagate to these
+ // nodes.
while (Val) {
if (Val & 1) {
if (Res.getNode())
N);
return nullptr;
}
- } else if (AI)
+ } else {
SDV = DAG.getDbgValue(Variable, Expression, N.getNode(), N.getResNo(),
true, 0, dl, SDNodeOrder);
- else {
- // Can't do anything with other non-AI cases yet.
- DEBUG(dbgs() << "Dropping debug info for " << DI << "\n");
- DEBUG(dbgs() << "non-AllocaInst issue for Address: \n\t");
- DEBUG(Address->dump());
- return nullptr;
}
DAG.AddDbgValue(SDV, N.getNode(), isParameter);
} else {
getValue(I.getArgOperand(1)),
getValue(I.getArgOperand(2))));
} else {
+ // TODO: Intrinsic calls should have fast-math-flags.
SDValue Mul = DAG.getNode(ISD::FMUL, sdl,
getValue(I.getArgOperand(0)).getValueType(),
getValue(I.getArgOperand(0)),
std::pair<SDValue, SDValue>
SelectionDAGBuilder::lowerInvokable(TargetLowering::CallLoweringInfo &CLI,
- MachineBasicBlock *LandingPad) {
+ const BasicBlock *EHPadBB) {
MachineModuleInfo &MMI = DAG.getMachineFunction().getMMI();
MCSymbol *BeginLabel = nullptr;
- if (LandingPad) {
+ if (EHPadBB) {
// Insert a label before the invoke call to mark the try range. This can be
// used to detect deletion of the invoke via the MachineModuleInfo.
BeginLabel = MMI.getContext().createTempSymbol();
unsigned CallSiteIndex = MMI.getCurrentCallSite();
if (CallSiteIndex) {
MMI.setCallSiteBeginLabel(BeginLabel, CallSiteIndex);
- LPadToCallSiteMap[LandingPad].push_back(CallSiteIndex);
+ LPadToCallSiteMap[FuncInfo.MBBMap[EHPadBB]].push_back(CallSiteIndex);
// Now that the call site is handled, stop tracking it.
MMI.setCurrentCallSite(0);
DAG.setRoot(Result.second);
}
- if (LandingPad) {
+ if (EHPadBB) {
// Insert a label at the end of the invoke call to mark the try range. This
// can be used to detect deletion of the invoke via the MachineModuleInfo.
MCSymbol *EndLabel = MMI.getContext().createTempSymbol();
DAG.setRoot(DAG.getEHLabel(getCurSDLoc(), getRoot(), EndLabel));
// Inform MachineModuleInfo of range.
- MMI.addInvoke(LandingPad, BeginLabel, EndLabel);
+ if (MMI.hasEHFunclets()) {
+ WinEHFuncInfo &EHInfo =
+ MMI.getWinEHFuncInfo(DAG.getMachineFunction().getFunction());
+ EHInfo.addIPToStateRange(EHPadBB, BeginLabel, EndLabel);
+ } else {
+ MMI.addInvoke(FuncInfo.MBBMap[EHPadBB], BeginLabel, EndLabel);
+ }
}
return Result;
void SelectionDAGBuilder::LowerCallTo(ImmutableCallSite CS, SDValue Callee,
bool isTailCall,
- MachineBasicBlock *LandingPad) {
+ const BasicBlock *EHPadBB) {
PointerType *PT = cast<PointerType>(CS.getCalledValue()->getType());
FunctionType *FTy = cast<FunctionType>(PT->getElementType());
Type *RetTy = FTy->getReturnType();
CLI.setDebugLoc(getCurSDLoc()).setChain(getRoot())
.setCallee(RetTy, FTy, Callee, std::move(Args), CS)
.setTailCall(isTailCall);
- std::pair<SDValue,SDValue> Result = lowerInvokable(CLI, LandingPad);
+ std::pair<SDValue, SDValue> Result = lowerInvokable(CLI, EHPadBB);
if (Result.first.getNode())
setValue(CS.getInstruction(), Result.first);
SDISelAsmOperandInfo &Input = ConstraintOperands[OpInfo.MatchingInput];
if (OpInfo.ConstraintVT != Input.ConstraintVT) {
- const TargetRegisterInfo *TRI = DAG.getSubtarget().getRegisterInfo();
+ const TargetRegisterInfo *TRI = DAG.getSubtarget().getRegisterInfo();
std::pair<unsigned, const TargetRegisterClass *> MatchRC =
TLI.getRegForInlineAsmConstraint(TRI, OpInfo.ConstraintCode,
OpInfo.ConstraintVT);
/// This is a helper for lowering intrinsics that follow a target calling
/// convention or require stack pointer adjustment. Only a subset of the
/// intrinsic's operands need to participate in the calling convention.
-std::pair<SDValue, SDValue>
-SelectionDAGBuilder::lowerCallOperands(ImmutableCallSite CS, unsigned ArgIdx,
- unsigned NumArgs, SDValue Callee,
- Type *ReturnTy,
- MachineBasicBlock *LandingPad,
- bool IsPatchPoint) {
+std::pair<SDValue, SDValue> SelectionDAGBuilder::lowerCallOperands(
+ ImmutableCallSite CS, unsigned ArgIdx, unsigned NumArgs, SDValue Callee,
+ Type *ReturnTy, const BasicBlock *EHPadBB, bool IsPatchPoint) {
TargetLowering::ArgListTy Args;
Args.reserve(NumArgs);
.setCallee(CS.getCallingConv(), ReturnTy, Callee, std::move(Args), NumArgs)
.setDiscardResult(CS->use_empty()).setIsPatchPoint(IsPatchPoint);
- return lowerInvokable(CLI, LandingPad);
+ return lowerInvokable(CLI, EHPadBB);
}
/// \brief Add a stack map intrinsic call's live variable operands to a stackmap
/// \brief Lower llvm.experimental.patchpoint directly to its target opcode.
void SelectionDAGBuilder::visitPatchpoint(ImmutableCallSite CS,
- MachineBasicBlock *LandingPad) {
+ const BasicBlock *EHPadBB) {
// void|i64 @llvm.experimental.patchpoint.void|i64(i64 <id>,
// i32 <numBytes>,
// i8* <target>,
unsigned NumCallArgs = IsAnyRegCC ? 0 : NumArgs;
Type *ReturnTy =
IsAnyRegCC ? Type::getVoidTy(*DAG.getContext()) : CS->getType();
- std::pair<SDValue, SDValue> Result =
- lowerCallOperands(CS, NumMetaOpers, NumCallArgs, Callee, ReturnTy,
- LandingPad, true);
+ std::pair<SDValue, SDValue> Result = lowerCallOperands(
+ CS, NumMetaOpers, NumCallArgs, Callee, ReturnTy, EHPadBB, true);
SDNode *CallEnd = Result.second.getNode();
if (HasDef && (CallEnd->getOpcode() == ISD::CopyFromReg))
uint32_t JumpWeight = I->Weight;
uint32_t FallthroughWeight = UnhandledWeights;
- // If Fallthrough is a target of the jump table, we evenly distribute
- // the weight on the edge to Fallthrough to successors of CurMBB.
- // Also update the weight on the edge from JumpMBB to Fallthrough.
+ // If the default statement is a target of the jump table, we evenly
+ // distribute the default weight to successors of CurMBB. Also update
+ // the weight on the edge from JumpMBB to Fallthrough.
for (MachineBasicBlock::succ_iterator SI = JumpMBB->succ_begin(),
SE = JumpMBB->succ_end();
SI != SE; ++SI) {
- if (*SI == Fallthrough) {
+ if (*SI == DefaultMBB) {
JumpWeight += DefaultWeight / 2;
FallthroughWeight -= DefaultWeight / 2;
JumpMBB->setSuccWeight(SI, DefaultWeight / 2);
projects / oota-llvm.git / blobdiff