#include "llvm/Instructions.h"
#include "llvm/Intrinsics.h"
#include "llvm/IntrinsicInst.h"
+#include "llvm/Module.h"
#include "llvm/CodeGen/FastISel.h"
#include "llvm/CodeGen/GCStrategy.h"
#include "llvm/CodeGen/GCMetadata.h"
#include "llvm/CodeGen/MachineJumpTableInfo.h"
#include "llvm/CodeGen/MachineModuleInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/PseudoSourceValue.h"
#include "llvm/CodeGen/SelectionDAG.h"
+#include "llvm/CodeGen/DwarfWriter.h"
+#include "llvm/Analysis/DebugInfo.h"
#include "llvm/Target/TargetRegisterInfo.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Target/TargetFrameInfo.h"
#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetIntrinsicInfo.h"
#include "llvm/Target/TargetLowering.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/Support/Compiler.h"
+#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/MathExtras.h"
+#include "llvm/Support/raw_ostream.h"
#include <algorithm>
using namespace llvm;
cl::init(0));
/// ComputeLinearIndex - Given an LLVM IR aggregate type and a sequence
-/// insertvalue or extractvalue indices that identify a member, return
+/// of insertvalue or extractvalue indices that identify a member, return
/// the linearized index of the start of the member.
///
static unsigned ComputeLinearIndex(const TargetLowering &TLI, const Type *Ty,
return ComputeLinearIndex(TLI, *EI, Indices+1, IndicesEnd, CurIndex);
CurIndex = ComputeLinearIndex(TLI, *EI, 0, 0, CurIndex);
}
+ return CurIndex;
}
// Given an array type, recursively traverse the elements.
else if (const ArrayType *ATy = dyn_cast<ArrayType>(Ty)) {
return ComputeLinearIndex(TLI, EltTy, Indices+1, IndicesEnd, CurIndex);
CurIndex = ComputeLinearIndex(TLI, EltTy, 0, 0, CurIndex);
}
+ return CurIndex;
}
// We haven't found the type we're looking for, so keep searching.
return CurIndex + 1;
// Given an array type, recursively traverse the elements.
if (const ArrayType *ATy = dyn_cast<ArrayType>(Ty)) {
const Type *EltTy = ATy->getElementType();
- uint64_t EltSize = TLI.getTargetData()->getABITypeSize(EltTy);
+ uint64_t EltSize = TLI.getTargetData()->getTypePaddedSize(EltTy);
for (unsigned i = 0, e = ATy->getNumElements(); i != e; ++i)
ComputeValueVTs(TLI, EltTy, ValueVTs, Offsets,
StartingOffset + i * EltSize);
/// 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 VISIBILITY_HIDDEN RegsForValue {
/// TLI - The TargetLowering object.
///
/// may need be promoted or synthesized from one or more registers.
///
SmallVector<MVT, 4> ValueVTs;
-
+
/// RegVTs - The value types of the registers. This is the same size as
/// ValueVTs and it records, for each value, what the type of the assigned
/// register or registers are. (Individual values are never synthesized
/// it is necessary to have a separate record of the types.
///
SmallVector<MVT, 4> RegVTs;
-
+
/// Regs - This list holds the registers assigned to the values.
/// Each legal or promoted value requires one register, and each
/// expanded value requires multiple registers.
///
SmallVector<unsigned, 4> Regs;
-
+
RegsForValue() : TLI(0) {}
-
+
RegsForValue(const TargetLowering &tli,
- const SmallVector<unsigned, 4> ®s,
+ const SmallVector<unsigned, 4> ®s,
MVT regvt, MVT valuevt)
: TLI(&tli), ValueVTs(1, valuevt), RegVTs(1, regvt), Regs(regs) {}
RegsForValue(const TargetLowering &tli,
- const SmallVector<unsigned, 4> ®s,
+ const SmallVector<unsigned, 4> ®s,
const SmallVector<MVT, 4> ®vts,
const SmallVector<MVT, 4> &valuevts)
: TLI(&tli), ValueVTs(valuevts), RegVTs(regvts), Regs(regs) {}
Reg += NumRegs;
}
}
-
+
/// append - Add the specified values to this one.
void append(const RegsForValue &RHS) {
TLI = RHS.TLI;
RegVTs.append(RHS.RegVTs.begin(), RHS.RegVTs.end());
Regs.append(RHS.Regs.begin(), RHS.Regs.end());
}
-
-
+
+
/// getCopyFromRegs - Emit a series of CopyFromReg nodes that copies from
- /// this value and returns the result as a ValueVTs value. This uses
+ /// this value and returns the result as a ValueVTs value. This uses
/// Chain/Flag as the input and updates them for the output Chain/Flag.
/// If the Flag pointer is NULL, no flag is used.
- SDValue getCopyFromRegs(SelectionDAG &DAG,
+ SDValue getCopyFromRegs(SelectionDAG &DAG, DebugLoc dl,
SDValue &Chain, SDValue *Flag) const;
/// getCopyToRegs - Emit a series of CopyToReg nodes that copies the
- /// specified value into the registers specified by this object. This uses
+ /// specified value into the registers specified by this object. This uses
/// Chain/Flag as the input and updates them for the output Chain/Flag.
/// If the Flag pointer is NULL, no flag is used.
- void getCopyToRegs(SDValue Val, SelectionDAG &DAG,
+ void getCopyToRegs(SDValue Val, SelectionDAG &DAG, DebugLoc dl,
SDValue &Chain, SDValue *Flag) const;
-
+
/// AddInlineAsmOperands - Add this value to the specified inlineasm node
- /// operand list. This adds the code marker and includes the number of
+ /// operand list. This adds the code marker and includes the number of
/// values added into it.
void AddInlineAsmOperands(unsigned Code, SelectionDAG &DAG,
std::vector<SDValue> &Ops) const;
}
/// isUsedOutsideOfDefiningBlock - Return true if this instruction is used by
-/// PHI nodes or outside of the basic block that defines it, or used by a
+/// PHI nodes or outside of the basic block that defines it, or used by a
/// switch or atomic instruction, which may expand to multiple basic blocks.
static bool isUsedOutsideOfDefiningBlock(Instruction *I) {
if (isa<PHINode>(I)) return true;
static bool isOnlyUsedInEntryBlock(Argument *A, bool EnableFastISel) {
// With FastISel active, we may be splitting blocks, so force creation
// of virtual registers for all non-dead arguments.
- if (EnableFastISel)
+ // Don't force virtual registers for byval arguments though, because
+ // fast-isel can't handle those in all cases.
+ if (EnableFastISel && !A->hasByValAttr())
return A->use_empty();
BasicBlock *Entry = A->getParent()->begin();
}
void FunctionLoweringInfo::set(Function &fn, MachineFunction &mf,
+ SelectionDAG &DAG,
bool EnableFastISel) {
Fn = &fn;
MF = &mf;
if (AllocaInst *AI = dyn_cast<AllocaInst>(I))
if (ConstantInt *CUI = dyn_cast<ConstantInt>(AI->getArraySize())) {
const Type *Ty = AI->getAllocatedType();
- uint64_t TySize = TLI.getTargetData()->getABITypeSize(Ty);
- unsigned Align =
+ uint64_t TySize = TLI.getTargetData()->getTypePaddedSize(Ty);
+ unsigned Align =
std::max((unsigned)TLI.getTargetData()->getPrefTypeAlignment(Ty),
AI->getAlignment());
// Create Machine PHI nodes for LLVM PHI nodes, lowering them as
// appropriate.
PHINode *PN;
- for (BasicBlock::iterator I = BB->begin();(PN = dyn_cast<PHINode>(I)); ++I){
- if (PN->use_empty()) continue;
-
+ DebugLoc DL;
+ for (BasicBlock::iterator
+ I = BB->begin(), E = BB->end(); I != E; ++I) {
+ if (CallInst *CI = dyn_cast<CallInst>(I)) {
+ if (Function *F = CI->getCalledFunction()) {
+ switch (F->getIntrinsicID()) {
+ default: break;
+ case Intrinsic::dbg_stoppoint: {
+ DwarfWriter *DW = DAG.getDwarfWriter();
+ DbgStopPointInst *SPI = cast<DbgStopPointInst>(I);
+
+ if (DW && DW->ValidDebugInfo(SPI->getContext())) {
+ DICompileUnit CU(cast<GlobalVariable>(SPI->getContext()));
+ unsigned SrcFile = DW->RecordSource(CU.getDirectory(),
+ CU.getFilename());
+ unsigned idx = MF->getOrCreateDebugLocID(SrcFile,
+ SPI->getLine(),
+ SPI->getColumn());
+ DL = DebugLoc::get(idx);
+ }
+
+ break;
+ }
+ case Intrinsic::dbg_func_start: {
+ DwarfWriter *DW = DAG.getDwarfWriter();
+ if (DW) {
+ DbgFuncStartInst *FSI = cast<DbgFuncStartInst>(I);
+ Value *SP = FSI->getSubprogram();
+
+ if (DW->ValidDebugInfo(SP)) {
+ DISubprogram Subprogram(cast<GlobalVariable>(SP));
+ DICompileUnit CU(Subprogram.getCompileUnit());
+ unsigned SrcFile = DW->RecordSource(CU.getDirectory(),
+ CU.getFilename());
+ unsigned Line = Subprogram.getLineNumber();
+ DL = DebugLoc::get(MF->getOrCreateDebugLocID(SrcFile, Line, 0));
+ }
+ }
+
+ break;
+ }
+ }
+ }
+ }
+
+ PN = dyn_cast<PHINode>(I);
+ if (!PN || PN->use_empty()) continue;
+
unsigned PHIReg = ValueMap[PN];
assert(PHIReg && "PHI node does not have an assigned virtual register!");
unsigned NumRegisters = TLI.getNumRegisters(VT);
const TargetInstrInfo *TII = MF->getTarget().getInstrInfo();
for (unsigned i = 0; i != NumRegisters; ++i)
- BuildMI(MBB, TII->get(TargetInstrInfo::PHI), PHIReg+i);
+ BuildMI(MBB, DL, TII->get(TargetInstrInfo::PHI), PHIReg + i);
PHIReg += NumRegisters;
}
}
/// larger then ValueVT then AssertOp can be used to specify whether the extra
/// bits are known to be zero (ISD::AssertZext) or sign extended from ValueVT
/// (ISD::AssertSext).
-static SDValue getCopyFromParts(SelectionDAG &DAG,
- const SDValue *Parts,
- unsigned NumParts,
- MVT PartVT,
- MVT ValueVT,
- ISD::NodeType AssertOp = ISD::DELETED_NODE) {
+static SDValue getCopyFromParts(SelectionDAG &DAG, DebugLoc dl,
+ const SDValue *Parts,
+ unsigned NumParts, MVT PartVT, MVT ValueVT,
+ ISD::NodeType AssertOp = ISD::DELETED_NODE) {
assert(NumParts > 0 && "No parts to assemble!");
- TargetLowering &TLI = DAG.getTargetLoweringInfo();
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
SDValue Val = Parts[0];
if (NumParts > 1) {
ValueVT : MVT::getIntegerVT(RoundBits);
SDValue Lo, Hi;
+ MVT HalfVT = ValueVT.isInteger() ?
+ MVT::getIntegerVT(RoundBits/2) :
+ MVT::getFloatingPointVT(RoundBits/2);
+
if (RoundParts > 2) {
- MVT HalfVT = MVT::getIntegerVT(RoundBits/2);
- Lo = getCopyFromParts(DAG, Parts, RoundParts/2, PartVT, HalfVT);
- Hi = getCopyFromParts(DAG, Parts+RoundParts/2, RoundParts/2,
+ Lo = getCopyFromParts(DAG, dl, Parts, RoundParts/2, PartVT, HalfVT);
+ Hi = getCopyFromParts(DAG, dl, Parts+RoundParts/2, RoundParts/2,
PartVT, HalfVT);
} else {
- Lo = Parts[0];
- Hi = Parts[1];
+ Lo = DAG.getNode(ISD::BIT_CONVERT, dl, HalfVT, Parts[0]);
+ Hi = DAG.getNode(ISD::BIT_CONVERT, dl, HalfVT, Parts[1]);
}
if (TLI.isBigEndian())
std::swap(Lo, Hi);
- Val = DAG.getNode(ISD::BUILD_PAIR, RoundVT, Lo, Hi);
+ Val = DAG.getNode(ISD::BUILD_PAIR, dl, RoundVT, Lo, Hi);
if (RoundParts < NumParts) {
// Assemble the trailing non-power-of-2 part.
unsigned OddParts = NumParts - RoundParts;
MVT OddVT = MVT::getIntegerVT(OddParts * PartBits);
- Hi = getCopyFromParts(DAG, Parts+RoundParts, OddParts, PartVT, OddVT);
+ Hi = getCopyFromParts(DAG, dl,
+ Parts+RoundParts, OddParts, PartVT, OddVT);
// Combine the round and odd parts.
Lo = Val;
if (TLI.isBigEndian())
std::swap(Lo, Hi);
MVT TotalVT = MVT::getIntegerVT(NumParts * PartBits);
- Hi = DAG.getNode(ISD::ANY_EXTEND, TotalVT, Hi);
- Hi = DAG.getNode(ISD::SHL, TotalVT, Hi,
+ Hi = DAG.getNode(ISD::ANY_EXTEND, dl, TotalVT, Hi);
+ Hi = DAG.getNode(ISD::SHL, dl, TotalVT, Hi,
DAG.getConstant(Lo.getValueType().getSizeInBits(),
- TLI.getShiftAmountTy()));
- Lo = DAG.getNode(ISD::ZERO_EXTEND, TotalVT, Lo);
- Val = DAG.getNode(ISD::OR, TotalVT, Lo, Hi);
+ TLI.getPointerTy()));
+ Lo = DAG.getNode(ISD::ZERO_EXTEND, dl, TotalVT, Lo);
+ Val = DAG.getNode(ISD::OR, dl, TotalVT, Lo, Hi);
}
} else {
// Handle a multi-element vector.
// If the register was not expanded, truncate or copy the value,
// as appropriate.
for (unsigned i = 0; i != NumParts; ++i)
- Ops[i] = getCopyFromParts(DAG, &Parts[i], 1,
+ Ops[i] = getCopyFromParts(DAG, dl, &Parts[i], 1,
PartVT, IntermediateVT);
} else if (NumParts > 0) {
// If the intermediate type was expanded, build the intermediate operands
"Must expand into a divisible number of parts!");
unsigned Factor = NumParts / NumIntermediates;
for (unsigned i = 0; i != NumIntermediates; ++i)
- Ops[i] = getCopyFromParts(DAG, &Parts[i * Factor], Factor,
+ Ops[i] = getCopyFromParts(DAG, dl, &Parts[i * Factor], Factor,
PartVT, IntermediateVT);
}
// Build a vector with BUILD_VECTOR or CONCAT_VECTORS from the intermediate
// operands.
Val = DAG.getNode(IntermediateVT.isVector() ?
- ISD::CONCAT_VECTORS : ISD::BUILD_VECTOR,
+ ISD::CONCAT_VECTORS : ISD::BUILD_VECTOR, dl,
ValueVT, &Ops[0], NumIntermediates);
}
}
if (PartVT.isVector()) {
assert(ValueVT.isVector() && "Unknown vector conversion!");
- return DAG.getNode(ISD::BIT_CONVERT, ValueVT, Val);
+ return DAG.getNode(ISD::BIT_CONVERT, dl, ValueVT, Val);
}
if (ValueVT.isVector()) {
assert(ValueVT.getVectorElementType() == PartVT &&
ValueVT.getVectorNumElements() == 1 &&
"Only trivial scalar-to-vector conversions should get here!");
- return DAG.getNode(ISD::BUILD_VECTOR, ValueVT, Val);
+ return DAG.getNode(ISD::BUILD_VECTOR, dl, ValueVT, Val);
}
if (PartVT.isInteger() &&
// indicate whether the truncated bits will always be
// zero or sign-extension.
if (AssertOp != ISD::DELETED_NODE)
- Val = DAG.getNode(AssertOp, PartVT, Val,
+ Val = DAG.getNode(AssertOp, dl, PartVT, Val,
DAG.getValueType(ValueVT));
- return DAG.getNode(ISD::TRUNCATE, ValueVT, Val);
+ return DAG.getNode(ISD::TRUNCATE, dl, ValueVT, Val);
} else {
- return DAG.getNode(ISD::ANY_EXTEND, ValueVT, Val);
+ return DAG.getNode(ISD::ANY_EXTEND, dl, ValueVT, Val);
}
}
if (PartVT.isFloatingPoint() && ValueVT.isFloatingPoint()) {
if (ValueVT.bitsLT(Val.getValueType()))
// FP_ROUND's are always exact here.
- return DAG.getNode(ISD::FP_ROUND, ValueVT, Val,
+ return DAG.getNode(ISD::FP_ROUND, dl, ValueVT, Val,
DAG.getIntPtrConstant(1));
- return DAG.getNode(ISD::FP_EXTEND, ValueVT, Val);
+ return DAG.getNode(ISD::FP_EXTEND, dl, ValueVT, Val);
}
if (PartVT.getSizeInBits() == ValueVT.getSizeInBits())
- return DAG.getNode(ISD::BIT_CONVERT, ValueVT, Val);
+ return DAG.getNode(ISD::BIT_CONVERT, dl, ValueVT, Val);
assert(0 && "Unknown mismatch!");
return SDValue();
/// getCopyToParts - Create a series of nodes that contain the specified value
/// split into legal parts. If the parts contain more bits than Val, then, for
/// integers, ExtendKind can be used to specify how to generate the extra bits.
-static void getCopyToParts(SelectionDAG &DAG,
- SDValue Val,
- SDValue *Parts,
- unsigned NumParts,
- MVT PartVT,
+static void getCopyToParts(SelectionDAG &DAG, DebugLoc dl, SDValue Val,
+ SDValue *Parts, unsigned NumParts, MVT PartVT,
ISD::NodeType ExtendKind = ISD::ANY_EXTEND) {
- TargetLowering &TLI = DAG.getTargetLoweringInfo();
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
MVT PtrVT = TLI.getPointerTy();
MVT ValueVT = Val.getValueType();
unsigned PartBits = PartVT.getSizeInBits();
// If the parts cover more bits than the value has, promote the value.
if (PartVT.isFloatingPoint() && ValueVT.isFloatingPoint()) {
assert(NumParts == 1 && "Do not know what to promote to!");
- Val = DAG.getNode(ISD::FP_EXTEND, PartVT, Val);
+ Val = DAG.getNode(ISD::FP_EXTEND, dl, PartVT, Val);
} else if (PartVT.isInteger() && ValueVT.isInteger()) {
ValueVT = MVT::getIntegerVT(NumParts * PartBits);
- Val = DAG.getNode(ExtendKind, ValueVT, Val);
+ Val = DAG.getNode(ExtendKind, dl, ValueVT, Val);
} else {
assert(0 && "Unknown mismatch!");
}
} else if (PartBits == ValueVT.getSizeInBits()) {
// Different types of the same size.
assert(NumParts == 1 && PartVT != ValueVT);
- Val = DAG.getNode(ISD::BIT_CONVERT, PartVT, Val);
+ Val = DAG.getNode(ISD::BIT_CONVERT, dl, PartVT, Val);
} else if (NumParts * PartBits < ValueVT.getSizeInBits()) {
// If the parts cover less bits than value has, truncate the value.
if (PartVT.isInteger() && ValueVT.isInteger()) {
ValueVT = MVT::getIntegerVT(NumParts * PartBits);
- Val = DAG.getNode(ISD::TRUNCATE, ValueVT, Val);
+ Val = DAG.getNode(ISD::TRUNCATE, dl, ValueVT, Val);
} else {
assert(0 && "Unknown mismatch!");
}
unsigned RoundParts = 1 << Log2_32(NumParts);
unsigned RoundBits = RoundParts * PartBits;
unsigned OddParts = NumParts - RoundParts;
- SDValue OddVal = DAG.getNode(ISD::SRL, ValueVT, Val,
- DAG.getConstant(RoundBits,
- TLI.getShiftAmountTy()));
- getCopyToParts(DAG, OddVal, Parts + RoundParts, OddParts, PartVT);
+ SDValue OddVal = DAG.getNode(ISD::SRL, dl, ValueVT, Val,
+ DAG.getConstant(RoundBits,
+ TLI.getPointerTy()));
+ getCopyToParts(DAG, dl, OddVal, Parts + RoundParts, OddParts, PartVT);
if (TLI.isBigEndian())
// The odd parts were reversed by getCopyToParts - unreverse them.
std::reverse(Parts + RoundParts, Parts + NumParts);
NumParts = RoundParts;
ValueVT = MVT::getIntegerVT(NumParts * PartBits);
- Val = DAG.getNode(ISD::TRUNCATE, ValueVT, Val);
+ Val = DAG.getNode(ISD::TRUNCATE, dl, ValueVT, Val);
}
// The number of parts is a power of 2. Repeatedly bisect the value using
// EXTRACT_ELEMENT.
- Parts[0] = DAG.getNode(ISD::BIT_CONVERT,
+ Parts[0] = DAG.getNode(ISD::BIT_CONVERT, dl,
MVT::getIntegerVT(ValueVT.getSizeInBits()),
Val);
for (unsigned StepSize = NumParts; StepSize > 1; StepSize /= 2) {
SDValue &Part0 = Parts[i];
SDValue &Part1 = Parts[i+StepSize/2];
- Part1 = DAG.getNode(ISD::EXTRACT_ELEMENT, ThisVT, Part0,
+ Part1 = DAG.getNode(ISD::EXTRACT_ELEMENT, dl,
+ ThisVT, Part0,
DAG.getConstant(1, PtrVT));
- Part0 = DAG.getNode(ISD::EXTRACT_ELEMENT, ThisVT, Part0,
+ Part0 = DAG.getNode(ISD::EXTRACT_ELEMENT, dl,
+ ThisVT, Part0,
DAG.getConstant(0, PtrVT));
if (ThisBits == PartBits && ThisVT != PartVT) {
- Part0 = DAG.getNode(ISD::BIT_CONVERT, PartVT, Part0);
- Part1 = DAG.getNode(ISD::BIT_CONVERT, PartVT, Part1);
+ Part0 = DAG.getNode(ISD::BIT_CONVERT, dl,
+ PartVT, Part0);
+ Part1 = DAG.getNode(ISD::BIT_CONVERT, dl,
+ PartVT, Part1);
}
}
}
if (NumParts == 1) {
if (PartVT != ValueVT) {
if (PartVT.isVector()) {
- Val = DAG.getNode(ISD::BIT_CONVERT, PartVT, Val);
+ Val = DAG.getNode(ISD::BIT_CONVERT, dl, PartVT, Val);
} else {
assert(ValueVT.getVectorElementType() == PartVT &&
ValueVT.getVectorNumElements() == 1 &&
"Only trivial vector-to-scalar conversions should get here!");
- Val = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, PartVT, Val,
+ Val = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl,
+ PartVT, Val,
DAG.getConstant(0, PtrVT));
}
}
// Handle a multi-element vector.
MVT IntermediateVT, RegisterVT;
unsigned NumIntermediates;
- unsigned NumRegs =
- DAG.getTargetLoweringInfo()
+ unsigned NumRegs = TLI
.getVectorTypeBreakdown(ValueVT, IntermediateVT, NumIntermediates,
RegisterVT);
unsigned NumElements = ValueVT.getVectorNumElements();
SmallVector<SDValue, 8> Ops(NumIntermediates);
for (unsigned i = 0; i != NumIntermediates; ++i)
if (IntermediateVT.isVector())
- Ops[i] = DAG.getNode(ISD::EXTRACT_SUBVECTOR,
+ Ops[i] = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl,
IntermediateVT, Val,
DAG.getConstant(i * (NumElements / NumIntermediates),
PtrVT));
else
- Ops[i] = DAG.getNode(ISD::EXTRACT_VECTOR_ELT,
- IntermediateVT, Val,
+ Ops[i] = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl,
+ IntermediateVT, Val,
DAG.getConstant(i, PtrVT));
// Split the intermediate operands into legal parts.
// If the register was not expanded, promote or copy the value,
// as appropriate.
for (unsigned i = 0; i != NumParts; ++i)
- getCopyToParts(DAG, Ops[i], &Parts[i], 1, PartVT);
+ getCopyToParts(DAG, dl, Ops[i], &Parts[i], 1, PartVT);
} else if (NumParts > 0) {
// If the intermediate type was expanded, split each the value into
// legal parts.
"Must expand into a divisible number of parts!");
unsigned Factor = NumParts / NumIntermediates;
for (unsigned i = 0; i != NumIntermediates; ++i)
- getCopyToParts(DAG, Ops[i], &Parts[i * Factor], Factor, PartVT);
+ getCopyToParts(DAG, dl, Ops[i], &Parts[i * Factor], Factor, PartVT);
}
}
}
// Otherwise, we have to make a token factor node.
- SDValue Root = DAG.getNode(ISD::TokenFactor, MVT::Other,
+ SDValue Root = DAG.getNode(ISD::TokenFactor, getCurDebugLoc(), MVT::Other,
&PendingLoads[0], PendingLoads.size());
PendingLoads.clear();
DAG.setRoot(Root);
PendingExports.push_back(Root);
}
- Root = DAG.getNode(ISD::TokenFactor, MVT::Other,
+ Root = DAG.getNode(ISD::TokenFactor, getCurDebugLoc(), MVT::Other,
&PendingExports[0],
PendingExports.size());
PendingExports.clear();
case Instruction::OPCODE:return visit##OPCODE((CLASS&)I);
#include "llvm/Instruction.def"
}
-}
+}
void SelectionDAGLowering::visitAdd(User &I) {
if (I.getType()->isFPOrFPVector())
SDValue SelectionDAGLowering::getValue(const Value *V) {
SDValue &N = NodeMap[V];
if (N.getNode()) return N;
-
+
if (Constant *C = const_cast<Constant*>(dyn_cast<Constant>(V))) {
MVT VT = TLI.getValueType(V->getType(), true);
-
+
if (ConstantInt *CI = dyn_cast<ConstantInt>(C))
return N = DAG.getConstant(*CI, VT);
if (GlobalValue *GV = dyn_cast<GlobalValue>(C))
return N = DAG.getGlobalAddress(GV, VT);
-
+
if (isa<ConstantPointerNull>(C))
return N = DAG.getConstant(0, TLI.getPointerTy());
-
+
if (ConstantFP *CFP = dyn_cast<ConstantFP>(C))
return N = DAG.getConstantFP(*CFP, VT);
-
+
if (isa<UndefValue>(C) && !isa<VectorType>(V->getType()) &&
!V->getType()->isAggregateType())
- return N = DAG.getNode(ISD::UNDEF, VT);
+ return N = DAG.getNode(ISD::UNDEF, getCurDebugLoc(), VT);
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C)) {
visit(CE->getOpcode(), *CE);
assert(N1.getNode() && "visit didn't populate the ValueMap!");
return N1;
}
-
+
if (isa<ConstantStruct>(C) || isa<ConstantArray>(C)) {
SmallVector<SDValue, 4> Constants;
for (User::const_op_iterator OI = C->op_begin(), OE = C->op_end();
for (unsigned i = 0; i != NumElts; ++i) {
MVT EltVT = ValueVTs[i];
if (isa<UndefValue>(C))
- Constants[i] = DAG.getNode(ISD::UNDEF, EltVT);
+ Constants[i] = DAG.getNode(ISD::UNDEF, getCurDebugLoc(), EltVT);
else if (EltVT.isFloatingPoint())
Constants[i] = DAG.getConstantFP(0, EltVT);
else
const VectorType *VecTy = cast<VectorType>(V->getType());
unsigned NumElements = VecTy->getNumElements();
-
+
// Now that we know the number and type of the elements, get that number of
// elements into the Ops array based on what kind of constant it is.
SmallVector<SDValue, 16> Ops;
SDValue Op;
if (isa<UndefValue>(C))
- Op = DAG.getNode(ISD::UNDEF, EltVT);
+ Op = DAG.getNode(ISD::UNDEF, getCurDebugLoc(), EltVT);
else if (EltVT.isFloatingPoint())
Op = DAG.getConstantFP(0, EltVT);
else
Op = DAG.getConstant(0, EltVT);
Ops.assign(NumElements, Op);
}
-
+
// Create a BUILD_VECTOR node.
- return NodeMap[V] = DAG.getNode(ISD::BUILD_VECTOR, VT, &Ops[0], Ops.size());
+ return NodeMap[V] = DAG.getNode(ISD::BUILD_VECTOR, getCurDebugLoc(),
+ VT, &Ops[0], Ops.size());
}
-
+
// If this is a static alloca, generate it as the frameindex instead of
// computation.
if (const AllocaInst *AI = dyn_cast<AllocaInst>(V)) {
if (SI != FuncInfo.StaticAllocaMap.end())
return DAG.getFrameIndex(SI->second, TLI.getPointerTy());
}
-
+
unsigned InReg = FuncInfo.ValueMap[V];
assert(InReg && "Value not in map!");
-
+
RegsForValue RFV(TLI, InReg, V->getType());
SDValue Chain = DAG.getEntryNode();
- return RFV.getCopyFromRegs(DAG, Chain, NULL);
+ return RFV.getCopyFromRegs(DAG, getCurDebugLoc(), Chain, NULL);
}
void SelectionDAGLowering::visitRet(ReturnInst &I) {
if (I.getNumOperands() == 0) {
- DAG.setRoot(DAG.getNode(ISD::RET, MVT::Other, getControlRoot()));
+ DAG.setRoot(DAG.getNode(ISD::RET, getCurDebugLoc(),
+ MVT::Other, getControlRoot()));
return;
}
-
+
SmallVector<SDValue, 8> NewValues;
NewValues.push_back(getControlRoot());
- for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i) {
- SDValue RetOp = getValue(I.getOperand(i));
-
+ for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i) {
SmallVector<MVT, 4> ValueVTs;
ComputeValueVTs(TLI, I.getOperand(i)->getType(), ValueVTs);
- for (unsigned j = 0, f = ValueVTs.size(); j != f; ++j) {
+ unsigned NumValues = ValueVTs.size();
+ if (NumValues == 0) continue;
+
+ SDValue RetOp = getValue(I.getOperand(i));
+ for (unsigned j = 0, f = NumValues; j != f; ++j) {
MVT VT = ValueVTs[j];
// 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.
+ // at least 32-bit. But this is not necessary for non-C calling
+ // conventions.
if (VT.isInteger()) {
MVT MinVT = TLI.getRegisterType(MVT::i32);
if (VT.bitsLT(MinVT))
MVT PartVT = TLI.getRegisterType(VT);
SmallVector<SDValue, 4> Parts(NumParts);
ISD::NodeType ExtendKind = ISD::ANY_EXTEND;
-
+
const Function *F = I.getParent()->getParent();
- if (F->paramHasAttr(0, ParamAttr::SExt))
+ if (F->paramHasAttr(0, Attribute::SExt))
ExtendKind = ISD::SIGN_EXTEND;
- else if (F->paramHasAttr(0, ParamAttr::ZExt))
+ else if (F->paramHasAttr(0, Attribute::ZExt))
ExtendKind = ISD::ZERO_EXTEND;
- getCopyToParts(DAG, SDValue(RetOp.getNode(), RetOp.getResNo() + j),
+ getCopyToParts(DAG, getCurDebugLoc(),
+ SDValue(RetOp.getNode(), RetOp.getResNo() + j),
&Parts[0], NumParts, PartVT, ExtendKind);
+ // 'inreg' on function refers to return value
+ ISD::ArgFlagsTy Flags = ISD::ArgFlagsTy();
+ if (F->paramHasAttr(0, Attribute::InReg))
+ Flags.setInReg();
for (unsigned i = 0; i < NumParts; ++i) {
NewValues.push_back(Parts[i]);
- NewValues.push_back(DAG.getArgFlags(ISD::ArgFlagsTy()));
+ NewValues.push_back(DAG.getArgFlags(Flags));
}
}
}
- DAG.setRoot(DAG.getNode(ISD::RET, MVT::Other,
+ DAG.setRoot(DAG.getNode(ISD::RET, getCurDebugLoc(), MVT::Other,
&NewValues[0], NewValues.size()));
}
void SelectionDAGLowering::ExportFromCurrentBlock(Value *V) {
// No need to export constants.
if (!isa<Instruction>(V) && !isa<Argument>(V)) return;
-
+
// Already exported?
if (FuncInfo.isExportedInst(V)) return;
// Can export from current BB.
if (VI->getParent() == FromBB)
return true;
-
+
// Is already exported, noop.
return FuncInfo.isExportedInst(V);
}
-
+
// If this is an argument, we can export it if the BB is the entry block or
// if it is already exported.
if (isa<Argument>(V)) {
// Otherwise, can only export this if it is already exported.
return FuncInfo.isExportedInst(V);
}
-
+
// Otherwise, constants can always be exported.
return true;
}
return true;
}
-/// FindMergedConditions - If Cond is an expression like
-void SelectionDAGLowering::FindMergedConditions(Value *Cond,
- MachineBasicBlock *TBB,
- MachineBasicBlock *FBB,
- MachineBasicBlock *CurBB,
- unsigned Opc) {
- // If this node is not part of the or/and tree, emit it as a branch.
- Instruction *BOp = dyn_cast<Instruction>(Cond);
+/// getFCmpCondCode - Return the ISD condition code corresponding to
+/// the given LLVM IR floating-point condition code. This includes
+/// consideration of global floating-point math flags.
+///
+static ISD::CondCode getFCmpCondCode(FCmpInst::Predicate Pred) {
+ ISD::CondCode FPC, FOC;
+ switch (Pred) {
+ case FCmpInst::FCMP_FALSE: FOC = FPC = ISD::SETFALSE; break;
+ case FCmpInst::FCMP_OEQ: FOC = ISD::SETEQ; FPC = ISD::SETOEQ; break;
+ case FCmpInst::FCMP_OGT: FOC = ISD::SETGT; FPC = ISD::SETOGT; break;
+ case FCmpInst::FCMP_OGE: FOC = ISD::SETGE; FPC = ISD::SETOGE; break;
+ case FCmpInst::FCMP_OLT: FOC = ISD::SETLT; FPC = ISD::SETOLT; break;
+ case FCmpInst::FCMP_OLE: FOC = ISD::SETLE; FPC = ISD::SETOLE; break;
+ case FCmpInst::FCMP_ONE: FOC = ISD::SETNE; FPC = ISD::SETONE; break;
+ case FCmpInst::FCMP_ORD: FOC = FPC = ISD::SETO; break;
+ case FCmpInst::FCMP_UNO: FOC = FPC = ISD::SETUO; break;
+ case FCmpInst::FCMP_UEQ: FOC = ISD::SETEQ; FPC = ISD::SETUEQ; break;
+ case FCmpInst::FCMP_UGT: FOC = ISD::SETGT; FPC = ISD::SETUGT; break;
+ case FCmpInst::FCMP_UGE: FOC = ISD::SETGE; FPC = ISD::SETUGE; break;
+ case FCmpInst::FCMP_ULT: FOC = ISD::SETLT; FPC = ISD::SETULT; break;
+ case FCmpInst::FCMP_ULE: FOC = ISD::SETLE; FPC = ISD::SETULE; break;
+ case FCmpInst::FCMP_UNE: FOC = ISD::SETNE; FPC = ISD::SETUNE; break;
+ case FCmpInst::FCMP_TRUE: FOC = FPC = ISD::SETTRUE; break;
+ default:
+ assert(0 && "Invalid FCmp predicate opcode!");
+ FOC = FPC = ISD::SETFALSE;
+ break;
+ }
+ if (FiniteOnlyFPMath())
+ return FOC;
+ else
+ return FPC;
+}
- if (!BOp || !(isa<BinaryOperator>(BOp) || isa<CmpInst>(BOp)) ||
- (unsigned)BOp->getOpcode() != Opc || !BOp->hasOneUse() ||
- BOp->getParent() != CurBB->getBasicBlock() ||
- !InBlock(BOp->getOperand(0), CurBB->getBasicBlock()) ||
- !InBlock(BOp->getOperand(1), CurBB->getBasicBlock())) {
- const BasicBlock *BB = CurBB->getBasicBlock();
-
- // If the leaf of the tree is a comparison, merge the condition into
- // the caseblock.
- if ((isa<ICmpInst>(Cond) || isa<FCmpInst>(Cond)) &&
- // The operands of the cmp have to be in this block. We don't know
- // how to export them from some other block. If this is the first block
- // of the sequence, no exporting is needed.
- (CurBB == CurMBB ||
- (isExportableFromCurrentBlock(BOp->getOperand(0), BB) &&
- isExportableFromCurrentBlock(BOp->getOperand(1), BB)))) {
- BOp = cast<Instruction>(Cond);
+/// getICmpCondCode - Return the ISD condition code corresponding to
+/// the given LLVM IR integer condition code.
+///
+static ISD::CondCode getICmpCondCode(ICmpInst::Predicate Pred) {
+ switch (Pred) {
+ case ICmpInst::ICMP_EQ: return ISD::SETEQ;
+ case ICmpInst::ICMP_NE: return ISD::SETNE;
+ case ICmpInst::ICMP_SLE: return ISD::SETLE;
+ case ICmpInst::ICMP_ULE: return ISD::SETULE;
+ case ICmpInst::ICMP_SGE: return ISD::SETGE;
+ case ICmpInst::ICMP_UGE: return ISD::SETUGE;
+ case ICmpInst::ICMP_SLT: return ISD::SETLT;
+ case ICmpInst::ICMP_ULT: return ISD::SETULT;
+ case ICmpInst::ICMP_SGT: return ISD::SETGT;
+ case ICmpInst::ICMP_UGT: return ISD::SETUGT;
+ default:
+ assert(0 && "Invalid ICmp predicate opcode!");
+ return ISD::SETNE;
+ }
+}
+
+/// EmitBranchForMergedCondition - Helper method for FindMergedConditions.
+/// This function emits a branch and is used at the leaves of an OR or an
+/// AND operator tree.
+///
+void
+SelectionDAGLowering::EmitBranchForMergedCondition(Value *Cond,
+ MachineBasicBlock *TBB,
+ MachineBasicBlock *FBB,
+ MachineBasicBlock *CurBB) {
+ const BasicBlock *BB = CurBB->getBasicBlock();
+
+ // If the leaf of the tree is a comparison, merge the condition into
+ // the caseblock.
+ if (CmpInst *BOp = dyn_cast<CmpInst>(Cond)) {
+ // The operands of the cmp have to be in this block. We don't know
+ // how to export them from some other block. If this is the first block
+ // of the sequence, no exporting is needed.
+ if (CurBB == CurMBB ||
+ (isExportableFromCurrentBlock(BOp->getOperand(0), BB) &&
+ isExportableFromCurrentBlock(BOp->getOperand(1), BB))) {
ISD::CondCode Condition;
if (ICmpInst *IC = dyn_cast<ICmpInst>(Cond)) {
- switch (IC->getPredicate()) {
- default: assert(0 && "Unknown icmp predicate opcode!");
- case ICmpInst::ICMP_EQ: Condition = ISD::SETEQ; break;
- case ICmpInst::ICMP_NE: Condition = ISD::SETNE; break;
- case ICmpInst::ICMP_SLE: Condition = ISD::SETLE; break;
- case ICmpInst::ICMP_ULE: Condition = ISD::SETULE; break;
- case ICmpInst::ICMP_SGE: Condition = ISD::SETGE; break;
- case ICmpInst::ICMP_UGE: Condition = ISD::SETUGE; break;
- case ICmpInst::ICMP_SLT: Condition = ISD::SETLT; break;
- case ICmpInst::ICMP_ULT: Condition = ISD::SETULT; break;
- case ICmpInst::ICMP_SGT: Condition = ISD::SETGT; break;
- case ICmpInst::ICMP_UGT: Condition = ISD::SETUGT; break;
- }
+ Condition = getICmpCondCode(IC->getPredicate());
} else if (FCmpInst *FC = dyn_cast<FCmpInst>(Cond)) {
- ISD::CondCode FPC, FOC;
- switch (FC->getPredicate()) {
- default: assert(0 && "Unknown fcmp predicate opcode!");
- case FCmpInst::FCMP_FALSE: FOC = FPC = ISD::SETFALSE; break;
- case FCmpInst::FCMP_OEQ: FOC = ISD::SETEQ; FPC = ISD::SETOEQ; break;
- case FCmpInst::FCMP_OGT: FOC = ISD::SETGT; FPC = ISD::SETOGT; break;
- case FCmpInst::FCMP_OGE: FOC = ISD::SETGE; FPC = ISD::SETOGE; break;
- case FCmpInst::FCMP_OLT: FOC = ISD::SETLT; FPC = ISD::SETOLT; break;
- case FCmpInst::FCMP_OLE: FOC = ISD::SETLE; FPC = ISD::SETOLE; break;
- case FCmpInst::FCMP_ONE: FOC = ISD::SETNE; FPC = ISD::SETONE; break;
- case FCmpInst::FCMP_ORD: FOC = FPC = ISD::SETO; break;
- case FCmpInst::FCMP_UNO: FOC = FPC = ISD::SETUO; break;
- case FCmpInst::FCMP_UEQ: FOC = ISD::SETEQ; FPC = ISD::SETUEQ; break;
- case FCmpInst::FCMP_UGT: FOC = ISD::SETGT; FPC = ISD::SETUGT; break;
- case FCmpInst::FCMP_UGE: FOC = ISD::SETGE; FPC = ISD::SETUGE; break;
- case FCmpInst::FCMP_ULT: FOC = ISD::SETLT; FPC = ISD::SETULT; break;
- case FCmpInst::FCMP_ULE: FOC = ISD::SETLE; FPC = ISD::SETULE; break;
- case FCmpInst::FCMP_UNE: FOC = ISD::SETNE; FPC = ISD::SETUNE; break;
- case FCmpInst::FCMP_TRUE: FOC = FPC = ISD::SETTRUE; break;
- }
- if (FiniteOnlyFPMath())
- Condition = FOC;
- else
- Condition = FPC;
+ Condition = getFCmpCondCode(FC->getPredicate());
} else {
Condition = ISD::SETEQ; // silence warning.
assert(0 && "Unknown compare instruction");
}
-
- CaseBlock CB(Condition, BOp->getOperand(0),
+
+ CaseBlock CB(Condition, BOp->getOperand(0),
BOp->getOperand(1), NULL, TBB, FBB, CurBB);
SwitchCases.push_back(CB);
return;
}
-
- // Create a CaseBlock record representing this branch.
- CaseBlock CB(ISD::SETEQ, Cond, ConstantInt::getTrue(),
- NULL, TBB, FBB, CurBB);
- SwitchCases.push_back(CB);
+ }
+
+ // Create a CaseBlock record representing this branch.
+ CaseBlock CB(ISD::SETEQ, Cond, ConstantInt::getTrue(),
+ NULL, TBB, FBB, CurBB);
+ SwitchCases.push_back(CB);
+}
+
+/// FindMergedConditions - If Cond is an expression like
+void SelectionDAGLowering::FindMergedConditions(Value *Cond,
+ MachineBasicBlock *TBB,
+ MachineBasicBlock *FBB,
+ MachineBasicBlock *CurBB,
+ unsigned Opc) {
+ // If this node is not part of the or/and tree, emit it as a branch.
+ Instruction *BOp = dyn_cast<Instruction>(Cond);
+ if (!BOp || !(isa<BinaryOperator>(BOp) || isa<CmpInst>(BOp)) ||
+ (unsigned)BOp->getOpcode() != Opc || !BOp->hasOneUse() ||
+ BOp->getParent() != CurBB->getBasicBlock() ||
+ !InBlock(BOp->getOperand(0), CurBB->getBasicBlock()) ||
+ !InBlock(BOp->getOperand(1), CurBB->getBasicBlock())) {
+ EmitBranchForMergedCondition(Cond, TBB, FBB, CurBB);
return;
}
-
-
+
// Create TmpBB after CurBB.
MachineFunction::iterator BBI = CurBB;
MachineFunction &MF = DAG.getMachineFunction();
MachineBasicBlock *TmpBB = MF.CreateMachineBasicBlock(CurBB->getBasicBlock());
CurBB->getParent()->insert(++BBI, TmpBB);
-
+
if (Opc == Instruction::Or) {
// Codegen X | Y as:
// jmp_if_X TBB
// jmp_if_Y TBB
// jmp FBB
//
-
+
// Emit the LHS condition.
FindMergedConditions(BOp->getOperand(0), TBB, TmpBB, CurBB, Opc);
-
+
// Emit the RHS condition into TmpBB.
FindMergedConditions(BOp->getOperand(1), TBB, FBB, TmpBB, Opc);
} else {
// jmp FBB
//
// This requires creation of TmpBB after CurBB.
-
+
// Emit the LHS condition.
FindMergedConditions(BOp->getOperand(0), TmpBB, FBB, CurBB, Opc);
-
+
// Emit the RHS condition into TmpBB.
FindMergedConditions(BOp->getOperand(1), TBB, FBB, TmpBB, Opc);
}
/// If the set of cases should be emitted as a series of branches, return true.
/// If we should emit this as a bunch of and/or'd together conditions, return
/// false.
-bool
+bool
SelectionDAGLowering::ShouldEmitAsBranches(const std::vector<CaseBlock> &Cases){
if (Cases.size() != 2) return true;
-
+
// If this is two comparisons of the same values or'd or and'd together, they
// will get folded into a single comparison, so don't emit two blocks.
if ((Cases[0].CmpLHS == Cases[1].CmpLHS &&
Cases[0].CmpLHS == Cases[1].CmpRHS)) {
return false;
}
-
+
return true;
}
if (I.isUnconditional()) {
// Update machine-CFG edges.
CurMBB->addSuccessor(Succ0MBB);
-
+
// If this is not a fall-through branch, emit the branch.
if (Succ0MBB != NextBlock)
- DAG.setRoot(DAG.getNode(ISD::BR, MVT::Other, getControlRoot(),
+ DAG.setRoot(DAG.getNode(ISD::BR, getCurDebugLoc(),
+ MVT::Other, getControlRoot(),
DAG.getBasicBlock(Succ0MBB)));
return;
}
// this as a sequence of branches instead of setcc's with and/or operations.
// For example, instead of something like:
// cmp A, B
- // C = seteq
+ // C = seteq
// cmp D, E
- // F = setle
+ // F = setle
// or C, F
// jnz foo
// Emit:
// jle foo
//
if (BinaryOperator *BOp = dyn_cast<BinaryOperator>(CondVal)) {
- if (BOp->hasOneUse() &&
+ if (BOp->hasOneUse() &&
(BOp->getOpcode() == Instruction::And ||
BOp->getOpcode() == Instruction::Or)) {
FindMergedConditions(BOp, Succ0MBB, Succ1MBB, CurMBB, BOp->getOpcode());
// exported from this block, export them now. This block should always
// be the first entry.
assert(SwitchCases[0].ThisBB == CurMBB && "Unexpected lowering!");
-
+
// Allow some cases to be rejected.
if (ShouldEmitAsBranches(SwitchCases)) {
for (unsigned i = 1, e = SwitchCases.size(); i != e; ++i) {
ExportFromCurrentBlock(SwitchCases[i].CmpLHS);
ExportFromCurrentBlock(SwitchCases[i].CmpRHS);
}
-
+
// Emit the branch for this block.
visitSwitchCase(SwitchCases[0]);
SwitchCases.erase(SwitchCases.begin());
return;
}
-
+
// Okay, we decided not to do this, remove any inserted MBB's and clear
// SwitchCases.
for (unsigned i = 1, e = SwitchCases.size(); i != e; ++i)
CurMBB->getParent()->erase(SwitchCases[i].ThisBB);
-
+
SwitchCases.clear();
}
}
-
+
// Create a CaseBlock record representing this branch.
CaseBlock CB(ISD::SETEQ, CondVal, ConstantInt::getTrue(),
NULL, Succ0MBB, Succ1MBB, CurMBB);
void SelectionDAGLowering::visitSwitchCase(CaseBlock &CB) {
SDValue Cond;
SDValue CondLHS = getValue(CB.CmpLHS);
-
- // Build the setcc now.
+ DebugLoc dl = getCurDebugLoc();
+
+ // Build the setcc now.
if (CB.CmpMHS == NULL) {
// Fold "(X == true)" to X and "(X == false)" to !X to
// handle common cases produced by branch lowering.
Cond = CondLHS;
else if (CB.CmpRHS == ConstantInt::getFalse() && CB.CC == ISD::SETEQ) {
SDValue True = DAG.getConstant(1, CondLHS.getValueType());
- Cond = DAG.getNode(ISD::XOR, CondLHS.getValueType(), CondLHS, True);
+ Cond = DAG.getNode(ISD::XOR, dl, CondLHS.getValueType(), CondLHS, True);
} else
- Cond = DAG.getSetCC(MVT::i1, CondLHS, getValue(CB.CmpRHS), CB.CC);
+ Cond = DAG.getSetCC(dl, MVT::i1, CondLHS, getValue(CB.CmpRHS), CB.CC);
} else {
assert(CB.CC == ISD::SETLE && "Can handle only LE ranges now");
- uint64_t Low = cast<ConstantInt>(CB.CmpLHS)->getSExtValue();
- uint64_t High = cast<ConstantInt>(CB.CmpRHS)->getSExtValue();
+ const APInt& Low = cast<ConstantInt>(CB.CmpLHS)->getValue();
+ const APInt& High = cast<ConstantInt>(CB.CmpRHS)->getValue();
SDValue CmpOp = getValue(CB.CmpMHS);
MVT VT = CmpOp.getValueType();
if (cast<ConstantInt>(CB.CmpLHS)->isMinValue(true)) {
- Cond = DAG.getSetCC(MVT::i1, CmpOp, DAG.getConstant(High, VT), ISD::SETLE);
+ Cond = DAG.getSetCC(dl, MVT::i1, CmpOp, DAG.getConstant(High, VT),
+ ISD::SETLE);
} else {
- SDValue SUB = DAG.getNode(ISD::SUB, VT, CmpOp, DAG.getConstant(Low, VT));
- Cond = DAG.getSetCC(MVT::i1, SUB,
+ SDValue SUB = DAG.getNode(ISD::SUB, dl,
+ VT, CmpOp, DAG.getConstant(Low, VT));
+ Cond = DAG.getSetCC(dl, MVT::i1, SUB,
DAG.getConstant(High-Low, VT), ISD::SETULE);
}
}
-
+
// Update successor info
CurMBB->addSuccessor(CB.TrueBB);
CurMBB->addSuccessor(CB.FalseBB);
-
+
// Set NextBlock to be the MBB immediately after the current one, if any.
// This is used to avoid emitting unnecessary branches to the next block.
MachineBasicBlock *NextBlock = 0;
MachineFunction::iterator BBI = CurMBB;
if (++BBI != CurMBB->getParent()->end())
NextBlock = BBI;
-
+
// If the lhs block is the next block, invert the condition so that we can
// fall through to the lhs instead of the rhs block.
if (CB.TrueBB == NextBlock) {
std::swap(CB.TrueBB, CB.FalseBB);
SDValue True = DAG.getConstant(1, Cond.getValueType());
- Cond = DAG.getNode(ISD::XOR, Cond.getValueType(), Cond, True);
+ Cond = DAG.getNode(ISD::XOR, dl, Cond.getValueType(), Cond, True);
}
- SDValue BrCond = DAG.getNode(ISD::BRCOND, MVT::Other, getControlRoot(), Cond,
- DAG.getBasicBlock(CB.TrueBB));
-
+ SDValue BrCond = DAG.getNode(ISD::BRCOND, dl,
+ MVT::Other, getControlRoot(), Cond,
+ DAG.getBasicBlock(CB.TrueBB));
+
// If the branch was constant folded, fix up the CFG.
if (BrCond.getOpcode() == ISD::BR) {
CurMBB->removeSuccessor(CB.FalseBB);
DAG.setRoot(BrCond);
} else {
// Otherwise, go ahead and insert the false branch.
- if (BrCond == getControlRoot())
+ if (BrCond == getControlRoot())
CurMBB->removeSuccessor(CB.TrueBB);
-
+
if (CB.FalseBB == NextBlock)
DAG.setRoot(BrCond);
else
- DAG.setRoot(DAG.getNode(ISD::BR, MVT::Other, BrCond,
+ DAG.setRoot(DAG.getNode(ISD::BR, dl, MVT::Other, BrCond,
DAG.getBasicBlock(CB.FalseBB)));
}
}
// Emit the code for the jump table
assert(JT.Reg != -1U && "Should lower JT Header first!");
MVT PTy = TLI.getPointerTy();
- SDValue Index = DAG.getCopyFromReg(getControlRoot(), JT.Reg, PTy);
+ SDValue Index = DAG.getCopyFromReg(getControlRoot(), getCurDebugLoc(),
+ JT.Reg, PTy);
SDValue Table = DAG.getJumpTable(JT.JTI, PTy);
- DAG.setRoot(DAG.getNode(ISD::BR_JT, MVT::Other, Index.getValue(1),
+ DAG.setRoot(DAG.getNode(ISD::BR_JT, getCurDebugLoc(),
+ MVT::Other, Index.getValue(1),
Table, Index));
- return;
}
/// visitJumpTableHeader - This function emits necessary code to produce index
/// in the JumpTable from switch case.
void SelectionDAGLowering::visitJumpTableHeader(JumpTable &JT,
JumpTableHeader &JTH) {
- // Subtract the lowest switch case value from the value being switched on
- // and conditional branch to default mbb if the result is greater than the
+ // Subtract the lowest switch case value from the value being switched on and
+ // conditional branch to default mbb if the result is greater than the
// difference between smallest and largest cases.
SDValue SwitchOp = getValue(JTH.SValue);
MVT VT = SwitchOp.getValueType();
- SDValue SUB = DAG.getNode(ISD::SUB, VT, SwitchOp,
- DAG.getConstant(JTH.First, VT));
-
- // The SDNode we just created, which holds the value being switched on
- // minus the the smallest case value, needs to be copied to a virtual
- // register so it can be used as an index into the jump table in a
- // subsequent basic block. This value may be smaller or larger than the
- // target's pointer type, and therefore require extension or truncating.
+ SDValue SUB = DAG.getNode(ISD::SUB, getCurDebugLoc(), VT, SwitchOp,
+ DAG.getConstant(JTH.First, VT));
+
+ // The SDNode we just created, which holds the value being switched on minus
+ // the the smallest case value, needs to be copied to a virtual register so it
+ // can be used as an index into the jump table in a subsequent basic block.
+ // This value may be smaller or larger than the target's pointer type, and
+ // therefore require extension or truncating.
if (VT.bitsGT(TLI.getPointerTy()))
- SwitchOp = DAG.getNode(ISD::TRUNCATE, TLI.getPointerTy(), SUB);
+ SwitchOp = DAG.getNode(ISD::TRUNCATE, getCurDebugLoc(),
+ TLI.getPointerTy(), SUB);
else
- SwitchOp = DAG.getNode(ISD::ZERO_EXTEND, TLI.getPointerTy(), SUB);
-
+ SwitchOp = DAG.getNode(ISD::ZERO_EXTEND, getCurDebugLoc(),
+ TLI.getPointerTy(), SUB);
+
unsigned JumpTableReg = FuncInfo.MakeReg(TLI.getPointerTy());
- SDValue CopyTo = DAG.getCopyToReg(getControlRoot(), JumpTableReg, SwitchOp);
+ SDValue CopyTo = DAG.getCopyToReg(getControlRoot(), getCurDebugLoc(),
+ JumpTableReg, SwitchOp);
JT.Reg = JumpTableReg;
- // Emit the range check for the jump table, and branch to the default
- // block for the switch statement if the value being switched on exceeds
- // the largest case in the switch.
- SDValue CMP = DAG.getSetCC(TLI.getSetCCResultType(SUB), SUB,
- DAG.getConstant(JTH.Last-JTH.First,VT),
- ISD::SETUGT);
+ // Emit the range check for the jump table, and branch to the default block
+ // for the switch statement if the value being switched on exceeds the largest
+ // case in the switch.
+ SDValue CMP = DAG.getSetCC(getCurDebugLoc(),
+ TLI.getSetCCResultType(SUB.getValueType()), SUB,
+ DAG.getConstant(JTH.Last-JTH.First,VT),
+ ISD::SETUGT);
// Set NextBlock to be the MBB immediately after the current one, if any.
// This is used to avoid emitting unnecessary branches to the next block.
if (++BBI != CurMBB->getParent()->end())
NextBlock = BBI;
- SDValue BrCond = DAG.getNode(ISD::BRCOND, MVT::Other, CopyTo, CMP,
- DAG.getBasicBlock(JT.Default));
+ SDValue BrCond = DAG.getNode(ISD::BRCOND, getCurDebugLoc(),
+ MVT::Other, CopyTo, CMP,
+ DAG.getBasicBlock(JT.Default));
if (JT.MBB == NextBlock)
DAG.setRoot(BrCond);
else
- DAG.setRoot(DAG.getNode(ISD::BR, MVT::Other, BrCond,
+ DAG.setRoot(DAG.getNode(ISD::BR, getCurDebugLoc(), MVT::Other, BrCond,
DAG.getBasicBlock(JT.MBB)));
-
- return;
}
/// visitBitTestHeader - This function emits necessary code to produce value
// Subtract the minimum value
SDValue SwitchOp = getValue(B.SValue);
MVT VT = SwitchOp.getValueType();
- SDValue SUB = DAG.getNode(ISD::SUB, VT, SwitchOp,
- DAG.getConstant(B.First, VT));
+ SDValue SUB = DAG.getNode(ISD::SUB, getCurDebugLoc(), VT, SwitchOp,
+ DAG.getConstant(B.First, VT));
// Check range
- SDValue RangeCmp = DAG.getSetCC(TLI.getSetCCResultType(SUB), SUB,
- DAG.getConstant(B.Range, VT),
- ISD::SETUGT);
+ SDValue RangeCmp = DAG.getSetCC(getCurDebugLoc(),
+ TLI.getSetCCResultType(SUB.getValueType()),
+ SUB, DAG.getConstant(B.Range, VT),
+ ISD::SETUGT);
SDValue ShiftOp;
- if (VT.bitsGT(TLI.getShiftAmountTy()))
- ShiftOp = DAG.getNode(ISD::TRUNCATE, TLI.getShiftAmountTy(), SUB);
+ if (VT.bitsGT(TLI.getPointerTy()))
+ ShiftOp = DAG.getNode(ISD::TRUNCATE, getCurDebugLoc(),
+ TLI.getPointerTy(), SUB);
else
- ShiftOp = DAG.getNode(ISD::ZERO_EXTEND, TLI.getShiftAmountTy(), SUB);
-
- // Make desired shift
- SDValue SwitchVal = DAG.getNode(ISD::SHL, TLI.getPointerTy(),
- DAG.getConstant(1, TLI.getPointerTy()),
- ShiftOp);
+ ShiftOp = DAG.getNode(ISD::ZERO_EXTEND, getCurDebugLoc(),
+ TLI.getPointerTy(), SUB);
- unsigned SwitchReg = FuncInfo.MakeReg(TLI.getPointerTy());
- SDValue CopyTo = DAG.getCopyToReg(getControlRoot(), SwitchReg, SwitchVal);
- B.Reg = SwitchReg;
+ B.Reg = FuncInfo.MakeReg(TLI.getPointerTy());
+ SDValue CopyTo = DAG.getCopyToReg(getControlRoot(), getCurDebugLoc(),
+ B.Reg, ShiftOp);
// Set NextBlock to be the MBB immediately after the current one, if any.
// This is used to avoid emitting unnecessary branches to the next block.
CurMBB->addSuccessor(B.Default);
CurMBB->addSuccessor(MBB);
- SDValue BrRange = DAG.getNode(ISD::BRCOND, MVT::Other, CopyTo, RangeCmp,
- DAG.getBasicBlock(B.Default));
-
+ SDValue BrRange = DAG.getNode(ISD::BRCOND, getCurDebugLoc(),
+ MVT::Other, CopyTo, RangeCmp,
+ DAG.getBasicBlock(B.Default));
+
if (MBB == NextBlock)
DAG.setRoot(BrRange);
else
- DAG.setRoot(DAG.getNode(ISD::BR, MVT::Other, CopyTo,
+ DAG.setRoot(DAG.getNode(ISD::BR, getCurDebugLoc(), MVT::Other, CopyTo,
DAG.getBasicBlock(MBB)));
-
- return;
}
/// visitBitTestCase - this function produces one "bit test"
void SelectionDAGLowering::visitBitTestCase(MachineBasicBlock* NextMBB,
unsigned Reg,
BitTestCase &B) {
+ // Make desired shift
+ SDValue ShiftOp = DAG.getCopyFromReg(getControlRoot(), getCurDebugLoc(), Reg,
+ TLI.getPointerTy());
+ SDValue SwitchVal = DAG.getNode(ISD::SHL, getCurDebugLoc(),
+ TLI.getPointerTy(),
+ DAG.getConstant(1, TLI.getPointerTy()),
+ ShiftOp);
+
// Emit bit tests and jumps
- SDValue SwitchVal = DAG.getCopyFromReg(getControlRoot(), Reg,
- TLI.getPointerTy());
-
- SDValue AndOp = DAG.getNode(ISD::AND, TLI.getPointerTy(), SwitchVal,
- DAG.getConstant(B.Mask, TLI.getPointerTy()));
- SDValue AndCmp = DAG.getSetCC(TLI.getSetCCResultType(AndOp), AndOp,
- DAG.getConstant(0, TLI.getPointerTy()),
- ISD::SETNE);
+ SDValue AndOp = DAG.getNode(ISD::AND, getCurDebugLoc(),
+ TLI.getPointerTy(), SwitchVal,
+ DAG.getConstant(B.Mask, TLI.getPointerTy()));
+ SDValue AndCmp = DAG.getSetCC(getCurDebugLoc(),
+ TLI.getSetCCResultType(AndOp.getValueType()),
+ AndOp, DAG.getConstant(0, TLI.getPointerTy()),
+ ISD::SETNE);
CurMBB->addSuccessor(B.TargetBB);
CurMBB->addSuccessor(NextMBB);
-
- SDValue BrAnd = DAG.getNode(ISD::BRCOND, MVT::Other, getControlRoot(),
- AndCmp, DAG.getBasicBlock(B.TargetBB));
+
+ SDValue BrAnd = DAG.getNode(ISD::BRCOND, getCurDebugLoc(),
+ MVT::Other, getControlRoot(),
+ AndCmp, DAG.getBasicBlock(B.TargetBB));
// Set NextBlock to be the MBB immediately after the current one, if any.
// This is used to avoid emitting unnecessary branches to the next block.
if (NextMBB == NextBlock)
DAG.setRoot(BrAnd);
else
- DAG.setRoot(DAG.getNode(ISD::BR, MVT::Other, BrAnd,
+ DAG.setRoot(DAG.getNode(ISD::BR, getCurDebugLoc(), MVT::Other, BrAnd,
DAG.getBasicBlock(NextMBB)));
-
- return;
}
void SelectionDAGLowering::visitInvoke(InvokeInst &I) {
MachineBasicBlock *Return = FuncInfo.MBBMap[I.getSuccessor(0)];
MachineBasicBlock *LandingPad = FuncInfo.MBBMap[I.getSuccessor(1)];
- if (isa<InlineAsm>(I.getCalledValue()))
+ const Value *Callee(I.getCalledValue());
+ if (isa<InlineAsm>(Callee))
visitInlineAsm(&I);
else
- LowerCallTo(&I, getValue(I.getOperand(0)), false, LandingPad);
+ LowerCallTo(&I, getValue(Callee), false, LandingPad);
// If the value of the invoke is used outside of its defining block, make it
// available as a virtual register.
CurMBB->addSuccessor(LandingPad);
// Drop into normal successor.
- DAG.setRoot(DAG.getNode(ISD::BR, MVT::Other, getControlRoot(),
+ DAG.setRoot(DAG.getNode(ISD::BR, getCurDebugLoc(),
+ MVT::Other, getControlRoot(),
DAG.getBasicBlock(Return)));
}
Value* SV,
MachineBasicBlock* Default) {
Case& BackCase = *(CR.Range.second-1);
-
+
// Size is the number of Cases represented by this range.
- unsigned Size = CR.Range.second - CR.Range.first;
+ size_t Size = CR.Range.second - CR.Range.first;
if (Size > 3)
- return false;
-
+ return false;
+
// Get the MachineFunction which holds the current MBB. This is used when
// inserting any additional MBBs necessary to represent the switch.
- MachineFunction *CurMF = CurMBB->getParent();
+ MachineFunction *CurMF = CurMBB->getParent();
// Figure out which block is immediately after the current one.
MachineBasicBlock *NextBlock = 0;
// is the same as the other, but has one bit unset that the other has set,
// use bit manipulation to do two compares at once. For example:
// "if (X == 6 || X == 4)" -> "if ((X|2) == 6)"
-
+
// Rearrange the case blocks so that the last one falls through if possible.
if (NextBlock && Default != NextBlock && BackCase.BB != NextBlock) {
// The last case block won't fall through into 'NextBlock' if we emit the
}
}
}
-
+
// Create a CaseBlock record representing a conditional branch to
// the Case's target mbb if the value being switched on SV is equal
// to C.
LHS = I->Low; MHS = SV; RHS = I->High;
}
CaseBlock CB(CC, LHS, RHS, MHS, I->BB, FallThrough, CurBlock);
-
+
// If emitting the first comparison, just call visitSwitchCase to emit the
// code into the current block. Otherwise, push the CaseBlock onto the
// vector to be later processed by SDISel, and insert the node's MBB
visitSwitchCase(CB);
else
SwitchCases.push_back(CB);
-
+
CurBlock = FallThrough;
}
static inline bool areJTsAllowed(const TargetLowering &TLI) {
return !DisableJumpTables &&
- (TLI.isOperationLegal(ISD::BR_JT, MVT::Other) ||
- TLI.isOperationLegal(ISD::BRIND, MVT::Other));
+ (TLI.isOperationLegalOrCustom(ISD::BR_JT, MVT::Other) ||
+ TLI.isOperationLegalOrCustom(ISD::BRIND, MVT::Other));
+}
+
+static APInt ComputeRange(const APInt &First, const APInt &Last) {
+ APInt LastExt(Last), FirstExt(First);
+ uint32_t BitWidth = std::max(Last.getBitWidth(), First.getBitWidth()) + 1;
+ LastExt.sext(BitWidth); FirstExt.sext(BitWidth);
+ return (LastExt - FirstExt + 1ULL);
}
-
+
/// handleJTSwitchCase - Emit jumptable for current switch case range
bool SelectionDAGLowering::handleJTSwitchCase(CaseRec& CR,
CaseRecVector& WorkList,
Case& FrontCase = *CR.Range.first;
Case& BackCase = *(CR.Range.second-1);
- int64_t First = cast<ConstantInt>(FrontCase.Low)->getSExtValue();
- int64_t Last = cast<ConstantInt>(BackCase.High)->getSExtValue();
+ const APInt& First = cast<ConstantInt>(FrontCase.Low)->getValue();
+ const APInt& Last = cast<ConstantInt>(BackCase.High)->getValue();
- uint64_t TSize = 0;
+ size_t TSize = 0;
for (CaseItr I = CR.Range.first, E = CR.Range.second;
I!=E; ++I)
TSize += I->size();
if (!areJTsAllowed(TLI) || TSize <= 3)
return false;
-
- double Density = (double)TSize / (double)((Last - First) + 1ULL);
+
+ APInt Range = ComputeRange(First, Last);
+ double Density = (double)TSize / Range.roundToDouble();
if (Density < 0.4)
return false;
- DOUT << "Lowering jump table\n"
- << "First entry: " << First << ". Last entry: " << Last << "\n"
- << "Size: " << TSize << ". Density: " << Density << "\n\n";
+ DEBUG(errs() << "Lowering jump table\n"
+ << "First entry: " << First << ". Last entry: " << Last << '\n'
+ << "Range: " << Range
+ << "Size: " << TSize << ". Density: " << Density << "\n\n");
// Get the MachineFunction which holds the current MBB. This is used when
// inserting any additional MBBs necessary to represent the switch.
CurMF->insert(BBI, JumpTableBB);
CR.CaseBB->addSuccessor(Default);
CR.CaseBB->addSuccessor(JumpTableBB);
-
+
// Build a vector of destination BBs, corresponding to each target
// of the jump table. If the value of the jump table slot corresponds to
// a case statement, push the case's BB onto the vector, otherwise, push
// the default BB.
std::vector<MachineBasicBlock*> DestBBs;
- int64_t TEI = First;
+ APInt TEI = First;
for (CaseItr I = CR.Range.first, E = CR.Range.second; I != E; ++TEI) {
- int64_t Low = cast<ConstantInt>(I->Low)->getSExtValue();
- int64_t High = cast<ConstantInt>(I->High)->getSExtValue();
-
- if ((Low <= TEI) && (TEI <= High)) {
+ const APInt& Low = cast<ConstantInt>(I->Low)->getValue();
+ const APInt& High = cast<ConstantInt>(I->High)->getValue();
+
+ if (Low.sle(TEI) && TEI.sle(High)) {
DestBBs.push_back(I->BB);
if (TEI==High)
++I;
DestBBs.push_back(Default);
}
}
-
+
// Update successor info. Add one edge to each unique successor.
- BitVector SuccsHandled(CR.CaseBB->getParent()->getNumBlockIDs());
- for (std::vector<MachineBasicBlock*>::iterator I = DestBBs.begin(),
+ BitVector SuccsHandled(CR.CaseBB->getParent()->getNumBlockIDs());
+ for (std::vector<MachineBasicBlock*>::iterator I = DestBBs.begin(),
E = DestBBs.end(); I != E; ++I) {
if (!SuccsHandled[(*I)->getNumber()]) {
SuccsHandled[(*I)->getNumber()] = true;
JumpTableBB->addSuccessor(*I);
}
}
-
+
// Create a jump table index for this jump table, or return an existing
// one.
unsigned JTI = CurMF->getJumpTableInfo()->getJumpTableIndex(DestBBs);
-
+
// Set the jump table information so that we can codegen it as a second
// MachineBasicBlock
JumpTable JT(-1U, JTI, JumpTableBB, Default);
JumpTableHeader JTH(First, Last, SV, CR.CaseBB, (CR.CaseBB == CurMBB));
if (CR.CaseBB == CurMBB)
visitJumpTableHeader(JT, JTH);
-
+
JTCases.push_back(JumpTableBlock(JTH, JT));
return true;
MachineBasicBlock* Default) {
// Get the MachineFunction which holds the current MBB. This is used when
// inserting any additional MBBs necessary to represent the switch.
- MachineFunction *CurMF = CurMBB->getParent();
+ MachineFunction *CurMF = CurMBB->getParent();
// Figure out which block is immediately after the current one.
MachineBasicBlock *NextBlock = 0;
// Size is the number of Cases represented by this range.
unsigned Size = CR.Range.second - CR.Range.first;
- int64_t First = cast<ConstantInt>(FrontCase.Low)->getSExtValue();
- int64_t Last = cast<ConstantInt>(BackCase.High)->getSExtValue();
+ const APInt& First = cast<ConstantInt>(FrontCase.Low)->getValue();
+ const APInt& Last = cast<ConstantInt>(BackCase.High)->getValue();
double FMetric = 0;
CaseItr Pivot = CR.Range.first + Size/2;
// Select optimal pivot, maximizing sum density of LHS and RHS. This will
// (heuristically) allow us to emit JumpTable's later.
- uint64_t TSize = 0;
+ size_t TSize = 0;
for (CaseItr I = CR.Range.first, E = CR.Range.second;
I!=E; ++I)
TSize += I->size();
- uint64_t LSize = FrontCase.size();
- uint64_t RSize = TSize-LSize;
- DOUT << "Selecting best pivot: \n"
- << "First: " << First << ", Last: " << Last <<"\n"
- << "LSize: " << LSize << ", RSize: " << RSize << "\n";
+ size_t LSize = FrontCase.size();
+ size_t RSize = TSize-LSize;
+ DEBUG(errs() << "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;
J!=E; ++I, ++J) {
- int64_t LEnd = cast<ConstantInt>(I->High)->getSExtValue();
- int64_t RBegin = cast<ConstantInt>(J->Low)->getSExtValue();
- assert((RBegin-LEnd>=1) && "Invalid case distance");
- double LDensity = (double)LSize / (double)((LEnd - First) + 1ULL);
- double RDensity = (double)RSize / (double)((Last - RBegin) + 1ULL);
- double Metric = Log2_64(RBegin-LEnd)*(LDensity+RDensity);
+ const APInt& LEnd = cast<ConstantInt>(I->High)->getValue();
+ const APInt& RBegin = cast<ConstantInt>(J->Low)->getValue();
+ APInt Range = ComputeRange(LEnd, RBegin);
+ assert((Range - 2ULL).isNonNegative() &&
+ "Invalid case distance");
+ double LDensity = (double)LSize / (LEnd - First + 1ULL).roundToDouble();
+ double RDensity = (double)RSize / (Last - RBegin + 1ULL).roundToDouble();
+ double Metric = Range.logBase2()*(LDensity+RDensity);
// Should always split in some non-trivial place
- DOUT <<"=>Step\n"
- << "LEnd: " << LEnd << ", RBegin: " << RBegin << "\n"
- << "LDensity: " << LDensity << ", RDensity: " << RDensity << "\n"
- << "Metric: " << Metric << "\n";
+ DEBUG(errs() <<"=>Step\n"
+ << "LEnd: " << LEnd << ", RBegin: " << RBegin << '\n'
+ << "LDensity: " << LDensity
+ << ", RDensity: " << RDensity << '\n'
+ << "Metric: " << Metric << '\n');
if (FMetric < Metric) {
Pivot = J;
FMetric = Metric;
- DOUT << "Current metric set to: " << FMetric << "\n";
+ DEBUG(errs() << "Current metric set to: " << FMetric << '\n');
}
LSize += J->size();
} else {
Pivot = CR.Range.first + Size/2;
}
-
+
CaseRange LHSR(CR.Range.first, Pivot);
CaseRange RHSR(Pivot, CR.Range.second);
Constant *C = Pivot->Low;
MachineBasicBlock *FalseBB = 0, *TrueBB = 0;
-
+
// We know that we branch to the LHS if the Value being switched on is
- // less than the Pivot value, C. We use this to optimize our binary
+ // less than the Pivot value, C. We use this to optimize our binary
// tree a bit, by recognizing that if SV is greater than or equal to the
- // LHS's Case Value, and that Case Value is exactly one less than the
+ // LHS's Case Value, and that Case Value is exactly one less than the
// Pivot's Value, then we can branch directly to the LHS's Target,
// rather than creating a leaf node for it.
if ((LHSR.second - LHSR.first) == 1 &&
LHSR.first->High == CR.GE &&
- cast<ConstantInt>(C)->getSExtValue() ==
- (cast<ConstantInt>(CR.GE)->getSExtValue() + 1LL)) {
+ cast<ConstantInt>(C)->getValue() ==
+ (cast<ConstantInt>(CR.GE)->getValue() + 1LL)) {
TrueBB = LHSR.first->BB;
} else {
TrueBB = CurMF->CreateMachineBasicBlock(LLVMBB);
CurMF->insert(BBI, TrueBB);
WorkList.push_back(CaseRec(TrueBB, C, CR.GE, LHSR));
}
-
+
// Similar to the optimization above, if the Value being switched on is
// known to be less than the Constant CR.LT, and the current Case Value
// is CR.LT - 1, then we can branch directly to the target block for
// the current Case Value, rather than emitting a RHS leaf node for it.
if ((RHSR.second - RHSR.first) == 1 && CR.LT &&
- cast<ConstantInt>(RHSR.first->Low)->getSExtValue() ==
- (cast<ConstantInt>(CR.LT)->getSExtValue() - 1LL)) {
+ cast<ConstantInt>(RHSR.first->Low)->getValue() ==
+ (cast<ConstantInt>(CR.LT)->getValue() - 1LL)) {
FalseBB = RHSR.first->BB;
} else {
FalseBB = CurMF->CreateMachineBasicBlock(LLVMBB);
}
// Create a CaseBlock record representing a conditional branch to
- // the LHS node if the value being switched on SV is less than C.
+ // 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);
// Get the MachineFunction which holds the current MBB. This is used when
// inserting any additional MBBs necessary to represent the switch.
- MachineFunction *CurMF = CurMBB->getParent();
+ MachineFunction *CurMF = CurMBB->getParent();
- unsigned numCmps = 0;
+ size_t numCmps = 0;
for (CaseItr I = CR.Range.first, E = CR.Range.second;
I!=E; ++I) {
// Single case counts one, case range - two.
- if (I->Low == I->High)
- numCmps +=1;
- else
- numCmps +=2;
+ numCmps += (I->Low == I->High ? 1 : 2);
}
-
+
// Count unique destinations
SmallSet<MachineBasicBlock*, 4> Dests;
for (CaseItr I = CR.Range.first, E = CR.Range.second; I!=E; ++I) {
// Don't bother the code below, if there are too much unique destinations
return false;
}
- DOUT << "Total number of unique destinations: " << Dests.size() << "\n"
- << "Total number of comparisons: " << numCmps << "\n";
-
+ DEBUG(errs() << "Total number of unique destinations: " << Dests.size() << '\n'
+ << "Total number of comparisons: " << numCmps << '\n');
+
// Compute span of values.
- Constant* minValue = FrontCase.Low;
- Constant* maxValue = BackCase.High;
- uint64_t range = cast<ConstantInt>(maxValue)->getSExtValue() -
- cast<ConstantInt>(minValue)->getSExtValue();
- DOUT << "Compare range: " << range << "\n"
- << "Low bound: " << cast<ConstantInt>(minValue)->getSExtValue() << "\n"
- << "High bound: " << cast<ConstantInt>(maxValue)->getSExtValue() << "\n";
-
- if (range>=IntPtrBits ||
+ 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'
+ << "Low bound: " << minValue << '\n'
+ << "High bound: " << maxValue << '\n');
+
+ if (cmpRange.uge(APInt(cmpRange.getBitWidth(), IntPtrBits)) ||
(!(Dests.size() == 1 && numCmps >= 3) &&
!(Dests.size() == 2 && numCmps >= 5) &&
!(Dests.size() >= 3 && numCmps >= 6)))
return false;
-
- DOUT << "Emitting bit tests\n";
- int64_t lowBound = 0;
-
+
+ DEBUG(errs() << "Emitting bit tests\n");
+ APInt lowBound = APInt::getNullValue(cmpRange.getBitWidth());
+
// 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 (cast<ConstantInt>(minValue)->getSExtValue() >= 0 &&
- cast<ConstantInt>(maxValue)->getSExtValue() < IntPtrBits) {
- range = cast<ConstantInt>(maxValue)->getSExtValue();
+ if (minValue.isNonNegative() &&
+ maxValue.slt(APInt(maxValue.getBitWidth(), IntPtrBits))) {
+ cmpRange = maxValue;
} else {
- lowBound = cast<ConstantInt>(minValue)->getSExtValue();
+ lowBound = minValue;
}
-
+
CaseBitsVector CasesBits;
unsigned i, count = 0;
for (i = 0; i < count; ++i)
if (Dest == CasesBits[i].BB)
break;
-
+
if (i == count) {
assert((count < 3) && "Too much destinations to test!");
CasesBits.push_back(CaseBits(0, Dest, 0));
count++;
}
-
- uint64_t lo = cast<ConstantInt>(I->Low)->getSExtValue() - lowBound;
- uint64_t hi = cast<ConstantInt>(I->High)->getSExtValue() - lowBound;
-
+
+ const APInt& lowValue = cast<ConstantInt>(I->Low)->getValue();
+ const APInt& highValue = cast<ConstantInt>(I->High)->getValue();
+
+ uint64_t lo = (lowValue - lowBound).getZExtValue();
+ uint64_t hi = (highValue - lowBound).getZExtValue();
+
for (uint64_t j = lo; j <= hi; j++) {
CasesBits[i].Mask |= 1ULL << j;
CasesBits[i].Bits++;
}
-
+
}
std::sort(CasesBits.begin(), CasesBits.end(), CaseBitsCmp());
-
+
BitTestInfo BTC;
// Figure out which block is immediately after the current one.
const BasicBlock *LLVMBB = CR.CaseBB->getBasicBlock();
- DOUT << "Cases:\n";
+ DEBUG(errs() << "Cases:\n");
for (unsigned i = 0, e = CasesBits.size(); i!=e; ++i) {
- DOUT << "Mask: " << CasesBits[i].Mask << ", Bits: " << CasesBits[i].Bits
- << ", BB: " << CasesBits[i].BB << "\n";
+ DEBUG(errs() << "Mask: " << CasesBits[i].Mask
+ << ", Bits: " << CasesBits[i].Bits
+ << ", BB: " << CasesBits[i].BB << '\n');
MachineBasicBlock *CaseBB = CurMF->CreateMachineBasicBlock(LLVMBB);
CurMF->insert(BBI, CaseBB);
CaseBB,
CasesBits[i].BB));
}
-
- BitTestBlock BTB(lowBound, range, SV,
+
+ BitTestBlock BTB(lowBound, cmpRange, SV,
-1U, (CR.CaseBB == CurMBB),
CR.CaseBB, Default, BTC);
if (CR.CaseBB == CurMBB)
visitBitTestHeader(BTB);
-
+
BitTestCases.push_back(BTB);
return true;
/// Clusterify - Transform simple list of Cases into list of CaseRange's
-unsigned SelectionDAGLowering::Clusterify(CaseVector& Cases,
+size_t SelectionDAGLowering::Clusterify(CaseVector& Cases,
const SwitchInst& SI) {
- unsigned numCmps = 0;
+ size_t numCmps = 0;
// Start with "simple" cases
- for (unsigned i = 1; i < SI.getNumSuccessors(); ++i) {
+ for (size_t i = 1; i < SI.getNumSuccessors(); ++i) {
MachineBasicBlock *SMBB = FuncInfo.MBBMap[SI.getSuccessor(i)];
Cases.push_back(Case(SI.getSuccessorValue(i),
SI.getSuccessorValue(i),
std::sort(Cases.begin(), Cases.end(), CaseCmp());
// Merge case into clusters
- if (Cases.size()>=2)
+ 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=++(Cases.begin()); J!=Cases.end(); ) {
- int64_t nextValue = cast<ConstantInt>(J->Low)->getSExtValue();
- int64_t currentValue = cast<ConstantInt>(I->High)->getSExtValue();
+ for (CaseItr I = Cases.begin(), J = ++(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)) {
+ if ((nextValue - currentValue == 1) && (currentBB == nextBB)) {
I->High = J->High;
J = Cases.erase(J);
} else {
return numCmps;
}
-void SelectionDAGLowering::visitSwitch(SwitchInst &SI) {
+void SelectionDAGLowering::visitSwitch(SwitchInst &SI) {
// Figure out which block is immediately after the current one.
MachineBasicBlock *NextBlock = 0;
MachineFunction::iterator BBI = CurMBB;
// If this is not a fall-through branch, emit the branch.
CurMBB->addSuccessor(Default);
if (Default != NextBlock)
- DAG.setRoot(DAG.getNode(ISD::BR, MVT::Other, getControlRoot(),
+ DAG.setRoot(DAG.getNode(ISD::BR, getCurDebugLoc(),
+ MVT::Other, getControlRoot(),
DAG.getBasicBlock(Default)));
-
return;
}
-
+
// If there are any non-default case statements, create a vector of Cases
// representing each one, and sort the vector so that we can efficiently
// create a binary search tree from them.
CaseVector Cases;
- unsigned numCmps = Clusterify(Cases, SI);
- DOUT << "Clusterify finished. Total clusters: " << Cases.size()
- << ". Total compares: " << numCmps << "\n";
+ size_t numCmps = Clusterify(Cases, SI);
+ DEBUG(errs() << "Clusterify finished. Total clusters: " << Cases.size()
+ << ". Total compares: " << numCmps << '\n');
+ numCmps = 0;
// Get the Value to be switched on and default basic blocks, which will be
// inserted into CaseBlock records, representing basic blocks in the binary
if (handleBitTestsSwitchCase(CR, WorkList, SV, Default))
continue;
-
+
// If the range has few cases (two or less) emit a series of specific
// tests.
if (handleSmallSwitchRange(CR, WorkList, SV, Default))
continue;
-
- // If the switch has more than 5 blocks, and at least 40% dense, and the
- // target supports indirect branches, then emit a jump table rather than
+
+ // If the switch has more than 5 blocks, and at least 40% dense, and the
+ // target supports indirect branches, then emit a jump table rather than
// lowering the switch to a binary tree of conditional branches.
if (handleJTSwitchCase(CR, WorkList, SV, Default))
continue;
-
+
// Emit binary tree. We need to pick a pivot, and push left and right ranges
// onto the worklist. Leafs are handled via handleSmallSwitchRange() call.
handleBTSplitSwitchCase(CR, WorkList, SV, Default);
Constant *CNZ = ConstantVector::get(&NZ[0], NZ.size());
if (CV == CNZ) {
SDValue Op2 = getValue(I.getOperand(1));
- setValue(&I, DAG.getNode(ISD::FNEG, Op2.getValueType(), Op2));
+ setValue(&I, DAG.getNode(ISD::FNEG, getCurDebugLoc(),
+ Op2.getValueType(), Op2));
return;
}
}
if (ConstantFP *CFP = dyn_cast<ConstantFP>(I.getOperand(0)))
if (CFP->isExactlyValue(ConstantFP::getNegativeZero(Ty)->getValueAPF())) {
SDValue Op2 = getValue(I.getOperand(1));
- setValue(&I, DAG.getNode(ISD::FNEG, Op2.getValueType(), Op2));
+ setValue(&I, DAG.getNode(ISD::FNEG, getCurDebugLoc(),
+ Op2.getValueType(), Op2));
return;
}
}
void SelectionDAGLowering::visitBinary(User &I, unsigned OpCode) {
SDValue Op1 = getValue(I.getOperand(0));
SDValue Op2 = getValue(I.getOperand(1));
-
- setValue(&I, DAG.getNode(OpCode, Op1.getValueType(), Op1, Op2));
+
+ setValue(&I, DAG.getNode(OpCode, getCurDebugLoc(),
+ Op1.getValueType(), Op1, Op2));
}
void SelectionDAGLowering::visitShift(User &I, unsigned Opcode) {
SDValue Op1 = getValue(I.getOperand(0));
SDValue Op2 = getValue(I.getOperand(1));
if (!isa<VectorType>(I.getType())) {
- if (TLI.getShiftAmountTy().bitsLT(Op2.getValueType()))
- Op2 = DAG.getNode(ISD::TRUNCATE, TLI.getShiftAmountTy(), Op2);
- else if (TLI.getShiftAmountTy().bitsGT(Op2.getValueType()))
- Op2 = DAG.getNode(ISD::ANY_EXTEND, TLI.getShiftAmountTy(), Op2);
+ if (TLI.getPointerTy().bitsLT(Op2.getValueType()))
+ Op2 = DAG.getNode(ISD::TRUNCATE, getCurDebugLoc(),
+ TLI.getPointerTy(), Op2);
+ else if (TLI.getPointerTy().bitsGT(Op2.getValueType()))
+ Op2 = DAG.getNode(ISD::ANY_EXTEND, getCurDebugLoc(),
+ TLI.getPointerTy(), Op2);
}
-
- setValue(&I, DAG.getNode(Opcode, Op1.getValueType(), Op1, Op2));
+
+ setValue(&I, DAG.getNode(Opcode, getCurDebugLoc(),
+ Op1.getValueType(), Op1, Op2));
}
void SelectionDAGLowering::visitICmp(User &I) {
predicate = ICmpInst::Predicate(IC->getPredicate());
SDValue Op1 = getValue(I.getOperand(0));
SDValue Op2 = getValue(I.getOperand(1));
- ISD::CondCode Opcode;
- switch (predicate) {
- case ICmpInst::ICMP_EQ : Opcode = ISD::SETEQ; break;
- case ICmpInst::ICMP_NE : Opcode = ISD::SETNE; break;
- case ICmpInst::ICMP_UGT : Opcode = ISD::SETUGT; break;
- case ICmpInst::ICMP_UGE : Opcode = ISD::SETUGE; break;
- case ICmpInst::ICMP_ULT : Opcode = ISD::SETULT; break;
- case ICmpInst::ICMP_ULE : Opcode = ISD::SETULE; break;
- case ICmpInst::ICMP_SGT : Opcode = ISD::SETGT; break;
- case ICmpInst::ICMP_SGE : Opcode = ISD::SETGE; break;
- case ICmpInst::ICMP_SLT : Opcode = ISD::SETLT; break;
- case ICmpInst::ICMP_SLE : Opcode = ISD::SETLE; break;
- default:
- assert(!"Invalid ICmp predicate value");
- Opcode = ISD::SETEQ;
- break;
- }
- setValue(&I, DAG.getSetCC(MVT::i1, Op1, Op2, Opcode));
+ ISD::CondCode Opcode = getICmpCondCode(predicate);
+ setValue(&I, DAG.getSetCC(getCurDebugLoc(),MVT::i1, Op1, Op2, Opcode));
}
void SelectionDAGLowering::visitFCmp(User &I) {
predicate = FCmpInst::Predicate(FC->getPredicate());
SDValue Op1 = getValue(I.getOperand(0));
SDValue Op2 = getValue(I.getOperand(1));
- ISD::CondCode Condition, FOC, FPC;
- switch (predicate) {
- case FCmpInst::FCMP_FALSE: FOC = FPC = ISD::SETFALSE; break;
- case FCmpInst::FCMP_OEQ: FOC = ISD::SETEQ; FPC = ISD::SETOEQ; break;
- case FCmpInst::FCMP_OGT: FOC = ISD::SETGT; FPC = ISD::SETOGT; break;
- case FCmpInst::FCMP_OGE: FOC = ISD::SETGE; FPC = ISD::SETOGE; break;
- case FCmpInst::FCMP_OLT: FOC = ISD::SETLT; FPC = ISD::SETOLT; break;
- case FCmpInst::FCMP_OLE: FOC = ISD::SETLE; FPC = ISD::SETOLE; break;
- case FCmpInst::FCMP_ONE: FOC = ISD::SETNE; FPC = ISD::SETONE; break;
- case FCmpInst::FCMP_ORD: FOC = FPC = ISD::SETO; break;
- case FCmpInst::FCMP_UNO: FOC = FPC = ISD::SETUO; break;
- case FCmpInst::FCMP_UEQ: FOC = ISD::SETEQ; FPC = ISD::SETUEQ; break;
- case FCmpInst::FCMP_UGT: FOC = ISD::SETGT; FPC = ISD::SETUGT; break;
- case FCmpInst::FCMP_UGE: FOC = ISD::SETGE; FPC = ISD::SETUGE; break;
- case FCmpInst::FCMP_ULT: FOC = ISD::SETLT; FPC = ISD::SETULT; break;
- case FCmpInst::FCMP_ULE: FOC = ISD::SETLE; FPC = ISD::SETULE; break;
- case FCmpInst::FCMP_UNE: FOC = ISD::SETNE; FPC = ISD::SETUNE; break;
- case FCmpInst::FCMP_TRUE: FOC = FPC = ISD::SETTRUE; break;
- default:
- assert(!"Invalid FCmp predicate value");
- FOC = FPC = ISD::SETFALSE;
- break;
- }
- if (FiniteOnlyFPMath())
- Condition = FOC;
- else
- Condition = FPC;
- setValue(&I, DAG.getSetCC(MVT::i1, Op1, Op2, Condition));
+ ISD::CondCode Condition = getFCmpCondCode(predicate);
+ setValue(&I, DAG.getSetCC(getCurDebugLoc(), MVT::i1, Op1, Op2, Condition));
}
void SelectionDAGLowering::visitVICmp(User &I) {
predicate = ICmpInst::Predicate(IC->getPredicate());
SDValue Op1 = getValue(I.getOperand(0));
SDValue Op2 = getValue(I.getOperand(1));
- ISD::CondCode Opcode;
- switch (predicate) {
- case ICmpInst::ICMP_EQ : Opcode = ISD::SETEQ; break;
- case ICmpInst::ICMP_NE : Opcode = ISD::SETNE; break;
- case ICmpInst::ICMP_UGT : Opcode = ISD::SETUGT; break;
- case ICmpInst::ICMP_UGE : Opcode = ISD::SETUGE; break;
- case ICmpInst::ICMP_ULT : Opcode = ISD::SETULT; break;
- case ICmpInst::ICMP_ULE : Opcode = ISD::SETULE; break;
- case ICmpInst::ICMP_SGT : Opcode = ISD::SETGT; break;
- case ICmpInst::ICMP_SGE : Opcode = ISD::SETGE; break;
- case ICmpInst::ICMP_SLT : Opcode = ISD::SETLT; break;
- case ICmpInst::ICMP_SLE : Opcode = ISD::SETLE; break;
- default:
- assert(!"Invalid ICmp predicate value");
- Opcode = ISD::SETEQ;
- break;
- }
- setValue(&I, DAG.getVSetCC(Op1.getValueType(), Op1, Op2, Opcode));
+ ISD::CondCode Opcode = getICmpCondCode(predicate);
+ setValue(&I, DAG.getVSetCC(getCurDebugLoc(), Op1.getValueType(),
+ Op1, Op2, Opcode));
}
void SelectionDAGLowering::visitVFCmp(User &I) {
predicate = FCmpInst::Predicate(FC->getPredicate());
SDValue Op1 = getValue(I.getOperand(0));
SDValue Op2 = getValue(I.getOperand(1));
- ISD::CondCode Condition, FOC, FPC;
- switch (predicate) {
- case FCmpInst::FCMP_FALSE: FOC = FPC = ISD::SETFALSE; break;
- case FCmpInst::FCMP_OEQ: FOC = ISD::SETEQ; FPC = ISD::SETOEQ; break;
- case FCmpInst::FCMP_OGT: FOC = ISD::SETGT; FPC = ISD::SETOGT; break;
- case FCmpInst::FCMP_OGE: FOC = ISD::SETGE; FPC = ISD::SETOGE; break;
- case FCmpInst::FCMP_OLT: FOC = ISD::SETLT; FPC = ISD::SETOLT; break;
- case FCmpInst::FCMP_OLE: FOC = ISD::SETLE; FPC = ISD::SETOLE; break;
- case FCmpInst::FCMP_ONE: FOC = ISD::SETNE; FPC = ISD::SETONE; break;
- case FCmpInst::FCMP_ORD: FOC = FPC = ISD::SETO; break;
- case FCmpInst::FCMP_UNO: FOC = FPC = ISD::SETUO; break;
- case FCmpInst::FCMP_UEQ: FOC = ISD::SETEQ; FPC = ISD::SETUEQ; break;
- case FCmpInst::FCMP_UGT: FOC = ISD::SETGT; FPC = ISD::SETUGT; break;
- case FCmpInst::FCMP_UGE: FOC = ISD::SETGE; FPC = ISD::SETUGE; break;
- case FCmpInst::FCMP_ULT: FOC = ISD::SETLT; FPC = ISD::SETULT; break;
- case FCmpInst::FCMP_ULE: FOC = ISD::SETLE; FPC = ISD::SETULE; break;
- case FCmpInst::FCMP_UNE: FOC = ISD::SETNE; FPC = ISD::SETUNE; break;
- case FCmpInst::FCMP_TRUE: FOC = FPC = ISD::SETTRUE; break;
- default:
- assert(!"Invalid VFCmp predicate value");
- FOC = FPC = ISD::SETFALSE;
- break;
- }
- if (FiniteOnlyFPMath())
- Condition = FOC;
- else
- Condition = FPC;
-
+ ISD::CondCode Condition = getFCmpCondCode(predicate);
MVT DestVT = TLI.getValueType(I.getType());
-
- setValue(&I, DAG.getVSetCC(DestVT, Op1, Op2, Condition));
+
+ setValue(&I, DAG.getVSetCC(getCurDebugLoc(), DestVT, Op1, Op2, Condition));
}
void SelectionDAGLowering::visitSelect(User &I) {
- SDValue Cond = getValue(I.getOperand(0));
- SDValue TrueVal = getValue(I.getOperand(1));
- SDValue FalseVal = getValue(I.getOperand(2));
- setValue(&I, DAG.getNode(ISD::SELECT, TrueVal.getValueType(), Cond,
- TrueVal, FalseVal));
+ SmallVector<MVT, 4> ValueVTs;
+ ComputeValueVTs(TLI, I.getType(), ValueVTs);
+ unsigned NumValues = ValueVTs.size();
+ if (NumValues != 0) {
+ SmallVector<SDValue, 4> Values(NumValues);
+ SDValue Cond = getValue(I.getOperand(0));
+ SDValue TrueVal = getValue(I.getOperand(1));
+ SDValue FalseVal = getValue(I.getOperand(2));
+
+ for (unsigned i = 0; i != NumValues; ++i)
+ Values[i] = DAG.getNode(ISD::SELECT, getCurDebugLoc(),
+ TrueVal.getValueType(), Cond,
+ SDValue(TrueVal.getNode(), TrueVal.getResNo() + i),
+ SDValue(FalseVal.getNode(), FalseVal.getResNo() + i));
+
+ setValue(&I, DAG.getNode(ISD::MERGE_VALUES, getCurDebugLoc(),
+ DAG.getVTList(&ValueVTs[0], NumValues),
+ &Values[0], NumValues));
+ }
}
// TruncInst cannot be a no-op cast because sizeof(src) > sizeof(dest).
SDValue N = getValue(I.getOperand(0));
MVT DestVT = TLI.getValueType(I.getType());
- setValue(&I, DAG.getNode(ISD::TRUNCATE, DestVT, N));
+ setValue(&I, DAG.getNode(ISD::TRUNCATE, getCurDebugLoc(), DestVT, N));
}
void SelectionDAGLowering::visitZExt(User &I) {
// ZExt also can't be a cast to bool for same reason. So, nothing much to do
SDValue N = getValue(I.getOperand(0));
MVT DestVT = TLI.getValueType(I.getType());
- setValue(&I, DAG.getNode(ISD::ZERO_EXTEND, DestVT, N));
+ setValue(&I, DAG.getNode(ISD::ZERO_EXTEND, getCurDebugLoc(), DestVT, N));
}
void SelectionDAGLowering::visitSExt(User &I) {
// SExt also can't be a cast to bool for same reason. So, nothing much to do
SDValue N = getValue(I.getOperand(0));
MVT DestVT = TLI.getValueType(I.getType());
- setValue(&I, DAG.getNode(ISD::SIGN_EXTEND, DestVT, N));
+ setValue(&I, DAG.getNode(ISD::SIGN_EXTEND, getCurDebugLoc(), DestVT, N));
}
void SelectionDAGLowering::visitFPTrunc(User &I) {
// FPTrunc is never a no-op cast, no need to check
SDValue N = getValue(I.getOperand(0));
MVT DestVT = TLI.getValueType(I.getType());
- setValue(&I, DAG.getNode(ISD::FP_ROUND, DestVT, N, DAG.getIntPtrConstant(0)));
+ setValue(&I, DAG.getNode(ISD::FP_ROUND, getCurDebugLoc(),
+ DestVT, N, DAG.getIntPtrConstant(0)));
}
-void SelectionDAGLowering::visitFPExt(User &I){
+void SelectionDAGLowering::visitFPExt(User &I){
// FPTrunc is never a no-op cast, no need to check
SDValue N = getValue(I.getOperand(0));
MVT DestVT = TLI.getValueType(I.getType());
- setValue(&I, DAG.getNode(ISD::FP_EXTEND, DestVT, N));
+ setValue(&I, DAG.getNode(ISD::FP_EXTEND, getCurDebugLoc(), DestVT, N));
}
-void SelectionDAGLowering::visitFPToUI(User &I) {
+void SelectionDAGLowering::visitFPToUI(User &I) {
// FPToUI is never a no-op cast, no need to check
SDValue N = getValue(I.getOperand(0));
MVT DestVT = TLI.getValueType(I.getType());
- setValue(&I, DAG.getNode(ISD::FP_TO_UINT, DestVT, N));
+ setValue(&I, DAG.getNode(ISD::FP_TO_UINT, getCurDebugLoc(), DestVT, N));
}
void SelectionDAGLowering::visitFPToSI(User &I) {
// FPToSI is never a no-op cast, no need to check
SDValue N = getValue(I.getOperand(0));
MVT DestVT = TLI.getValueType(I.getType());
- setValue(&I, DAG.getNode(ISD::FP_TO_SINT, DestVT, N));
+ setValue(&I, DAG.getNode(ISD::FP_TO_SINT, getCurDebugLoc(), DestVT, N));
}
-void SelectionDAGLowering::visitUIToFP(User &I) {
+void SelectionDAGLowering::visitUIToFP(User &I) {
// UIToFP is never a no-op cast, no need to check
SDValue N = getValue(I.getOperand(0));
MVT DestVT = TLI.getValueType(I.getType());
- setValue(&I, DAG.getNode(ISD::UINT_TO_FP, DestVT, N));
+ setValue(&I, DAG.getNode(ISD::UINT_TO_FP, getCurDebugLoc(), DestVT, N));
}
-void SelectionDAGLowering::visitSIToFP(User &I){
- // UIToFP is never a no-op cast, no need to check
+void SelectionDAGLowering::visitSIToFP(User &I){
+ // SIToFP is never a no-op cast, no need to check
SDValue N = getValue(I.getOperand(0));
MVT DestVT = TLI.getValueType(I.getType());
- setValue(&I, DAG.getNode(ISD::SINT_TO_FP, DestVT, N));
+ setValue(&I, DAG.getNode(ISD::SINT_TO_FP, getCurDebugLoc(), DestVT, N));
}
void SelectionDAGLowering::visitPtrToInt(User &I) {
MVT DestVT = TLI.getValueType(I.getType());
SDValue Result;
if (DestVT.bitsLT(SrcVT))
- Result = DAG.getNode(ISD::TRUNCATE, DestVT, N);
- else
+ Result = DAG.getNode(ISD::TRUNCATE, getCurDebugLoc(), DestVT, N);
+ else
// Note: ZERO_EXTEND can handle cases where the sizes are equal too
- Result = DAG.getNode(ISD::ZERO_EXTEND, DestVT, N);
+ Result = DAG.getNode(ISD::ZERO_EXTEND, getCurDebugLoc(), DestVT, N);
setValue(&I, Result);
}
MVT SrcVT = N.getValueType();
MVT DestVT = TLI.getValueType(I.getType());
if (DestVT.bitsLT(SrcVT))
- setValue(&I, DAG.getNode(ISD::TRUNCATE, DestVT, N));
- else
+ setValue(&I, DAG.getNode(ISD::TRUNCATE, getCurDebugLoc(), DestVT, N));
+ else
// Note: ZERO_EXTEND can handle cases where the sizes are equal too
- setValue(&I, DAG.getNode(ISD::ZERO_EXTEND, DestVT, N));
+ setValue(&I, DAG.getNode(ISD::ZERO_EXTEND, getCurDebugLoc(),
+ DestVT, N));
}
-void SelectionDAGLowering::visitBitCast(User &I) {
+void SelectionDAGLowering::visitBitCast(User &I) {
SDValue N = getValue(I.getOperand(0));
MVT DestVT = TLI.getValueType(I.getType());
- // BitCast assures us that source and destination are the same size so this
+ // BitCast assures us that source and destination are the same size so this
// is either a BIT_CONVERT or a no-op.
if (DestVT != N.getValueType())
- setValue(&I, DAG.getNode(ISD::BIT_CONVERT, DestVT, N)); // convert types
+ setValue(&I, DAG.getNode(ISD::BIT_CONVERT, getCurDebugLoc(),
+ DestVT, N)); // convert types
else
setValue(&I, N); // noop cast.
}
void SelectionDAGLowering::visitInsertElement(User &I) {
SDValue InVec = getValue(I.getOperand(0));
SDValue InVal = getValue(I.getOperand(1));
- SDValue InIdx = DAG.getNode(ISD::ZERO_EXTEND, TLI.getPointerTy(),
+ SDValue InIdx = DAG.getNode(ISD::ZERO_EXTEND, getCurDebugLoc(),
+ TLI.getPointerTy(),
getValue(I.getOperand(2)));
- setValue(&I, DAG.getNode(ISD::INSERT_VECTOR_ELT,
+ setValue(&I, DAG.getNode(ISD::INSERT_VECTOR_ELT, getCurDebugLoc(),
TLI.getValueType(I.getType()),
InVec, InVal, InIdx));
}
void SelectionDAGLowering::visitExtractElement(User &I) {
SDValue InVec = getValue(I.getOperand(0));
- SDValue InIdx = DAG.getNode(ISD::ZERO_EXTEND, TLI.getPointerTy(),
+ SDValue InIdx = DAG.getNode(ISD::ZERO_EXTEND, getCurDebugLoc(),
+ TLI.getPointerTy(),
getValue(I.getOperand(1)));
- setValue(&I, DAG.getNode(ISD::EXTRACT_VECTOR_ELT,
+ setValue(&I, DAG.getNode(ISD::EXTRACT_VECTOR_ELT, getCurDebugLoc(),
TLI.getValueType(I.getType()), InVec, InIdx));
}
+
+// Utility for visitShuffleVector - Returns true if the mask is mask starting
+// from SIndx and increasing to the element length (undefs are allowed).
+static bool SequentialMask(SDValue Mask, unsigned SIndx) {
+ unsigned MaskNumElts = Mask.getNumOperands();
+ for (unsigned i = 0; i != MaskNumElts; ++i) {
+ if (Mask.getOperand(i).getOpcode() != ISD::UNDEF) {
+ unsigned Idx = cast<ConstantSDNode>(Mask.getOperand(i))->getZExtValue();
+ if (Idx != i + SIndx)
+ return false;
+ }
+ }
+ return true;
+}
+
void SelectionDAGLowering::visitShuffleVector(User &I) {
- SDValue V1 = getValue(I.getOperand(0));
- SDValue V2 = getValue(I.getOperand(1));
+ SDValue Src1 = getValue(I.getOperand(0));
+ SDValue Src2 = getValue(I.getOperand(1));
SDValue Mask = getValue(I.getOperand(2));
- setValue(&I, DAG.getNode(ISD::VECTOR_SHUFFLE,
- TLI.getValueType(I.getType()),
- V1, V2, Mask));
+ MVT VT = TLI.getValueType(I.getType());
+ MVT SrcVT = Src1.getValueType();
+ int MaskNumElts = Mask.getNumOperands();
+ int SrcNumElts = SrcVT.getVectorNumElements();
+
+ if (SrcNumElts == MaskNumElts) {
+ setValue(&I, DAG.getNode(ISD::VECTOR_SHUFFLE, getCurDebugLoc(),
+ VT, Src1, Src2, Mask));
+ return;
+ }
+
+ // Normalize the shuffle vector since mask and vector length don't match.
+ MVT MaskEltVT = Mask.getValueType().getVectorElementType();
+
+ if (SrcNumElts < MaskNumElts && MaskNumElts % SrcNumElts == 0) {
+ // Mask is longer than the source vectors and is a multiple of the source
+ // vectors. We can use concatenate vector to make the mask and vectors
+ // lengths match.
+ if (SrcNumElts*2 == MaskNumElts && SequentialMask(Mask, 0)) {
+ // The shuffle is concatenating two vectors together.
+ setValue(&I, DAG.getNode(ISD::CONCAT_VECTORS, getCurDebugLoc(),
+ VT, Src1, Src2));
+ return;
+ }
+
+ // Pad both vectors with undefs to make them the same length as the mask.
+ unsigned NumConcat = MaskNumElts / SrcNumElts;
+ SDValue UndefVal = DAG.getNode(ISD::UNDEF, getCurDebugLoc(), SrcVT);
+
+ SDValue* MOps1 = new SDValue[NumConcat];
+ SDValue* MOps2 = new SDValue[NumConcat];
+ MOps1[0] = Src1;
+ MOps2[0] = Src2;
+ for (unsigned i = 1; i != NumConcat; ++i) {
+ MOps1[i] = UndefVal;
+ MOps2[i] = UndefVal;
+ }
+ Src1 = DAG.getNode(ISD::CONCAT_VECTORS, getCurDebugLoc(),
+ VT, MOps1, NumConcat);
+ Src2 = DAG.getNode(ISD::CONCAT_VECTORS, getCurDebugLoc(),
+ VT, MOps2, NumConcat);
+
+ delete [] MOps1;
+ delete [] MOps2;
+
+ // Readjust mask for new input vector length.
+ SmallVector<SDValue, 8> MappedOps;
+ for (int i = 0; i != MaskNumElts; ++i) {
+ if (Mask.getOperand(i).getOpcode() == ISD::UNDEF) {
+ MappedOps.push_back(Mask.getOperand(i));
+ } else {
+ int Idx = cast<ConstantSDNode>(Mask.getOperand(i))->getZExtValue();
+ if (Idx < SrcNumElts)
+ MappedOps.push_back(DAG.getConstant(Idx, MaskEltVT));
+ else
+ MappedOps.push_back(DAG.getConstant(Idx + MaskNumElts - SrcNumElts,
+ MaskEltVT));
+ }
+ }
+ Mask = DAG.getNode(ISD::BUILD_VECTOR, getCurDebugLoc(),
+ Mask.getValueType(),
+ &MappedOps[0], MappedOps.size());
+
+ setValue(&I, DAG.getNode(ISD::VECTOR_SHUFFLE, getCurDebugLoc(),
+ VT, Src1, Src2, Mask));
+ return;
+ }
+
+ if (SrcNumElts > MaskNumElts) {
+ // Resulting vector is shorter than the incoming vector.
+ if (SrcNumElts == MaskNumElts && SequentialMask(Mask,0)) {
+ // Shuffle extracts 1st vector.
+ setValue(&I, Src1);
+ return;
+ }
+
+ if (SrcNumElts == MaskNumElts && SequentialMask(Mask,MaskNumElts)) {
+ // Shuffle extracts 2nd vector.
+ setValue(&I, Src2);
+ return;
+ }
+
+ // 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] = { SrcNumElts+1, SrcNumElts+1};
+ int MaxRange[2] = {-1, -1};
+
+ for (int i = 0; i != MaskNumElts; ++i) {
+ SDValue Arg = Mask.getOperand(i);
+ if (Arg.getOpcode() != ISD::UNDEF) {
+ assert(isa<ConstantSDNode>(Arg) && "Invalid VECTOR_SHUFFLE mask!");
+ int Idx = cast<ConstantSDNode>(Arg)->getZExtValue();
+ int Input = 0;
+ if (Idx >= SrcNumElts) {
+ Input = 1;
+ Idx -= SrcNumElts;
+ }
+ if (Idx > MaxRange[Input])
+ MaxRange[Input] = Idx;
+ if (Idx < MinRange[Input])
+ MinRange[Input] = Idx;
+ }
+ }
+
+ // 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 StartIdx[2]; // StartIdx to extract from
+ for (int Input=0; Input < 2; ++Input) {
+ if (MinRange[Input] == SrcNumElts+1 && MaxRange[Input] == -1) {
+ RangeUse[Input] = 0; // Unused
+ StartIdx[Input] = 0;
+ } else if (MaxRange[Input] - MinRange[Input] < 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] < 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] < MaskNumElts &&
+ StartIdx[Input] + MaskNumElts < SrcNumElts)
+ RangeUse[Input] = 1; // Extract from a multiple of the mask length.
+ }
+ }
+ }
+
+ if (RangeUse[0] == 0 && RangeUse[0] == 0) {
+ setValue(&I, DAG.getNode(ISD::UNDEF,
+ getCurDebugLoc(), VT)); // Vectors are not used.
+ return;
+ }
+ else if (RangeUse[0] < 2 && RangeUse[1] < 2) {
+ // Extract appropriate subvector and generate a vector shuffle
+ for (int Input=0; Input < 2; ++Input) {
+ SDValue& Src = Input == 0 ? Src1 : Src2;
+ if (RangeUse[Input] == 0) {
+ Src = DAG.getNode(ISD::UNDEF, getCurDebugLoc(), VT);
+ } else {
+ Src = DAG.getNode(ISD::EXTRACT_SUBVECTOR, getCurDebugLoc(), VT,
+ Src, DAG.getIntPtrConstant(StartIdx[Input]));
+ }
+ }
+ // Calculate new mask.
+ SmallVector<SDValue, 8> MappedOps;
+ for (int i = 0; i != MaskNumElts; ++i) {
+ SDValue Arg = Mask.getOperand(i);
+ if (Arg.getOpcode() == ISD::UNDEF) {
+ MappedOps.push_back(Arg);
+ } else {
+ int Idx = cast<ConstantSDNode>(Arg)->getZExtValue();
+ if (Idx < SrcNumElts)
+ MappedOps.push_back(DAG.getConstant(Idx - StartIdx[0], MaskEltVT));
+ else {
+ Idx = Idx - SrcNumElts - StartIdx[1] + MaskNumElts;
+ MappedOps.push_back(DAG.getConstant(Idx, MaskEltVT));
+ }
+ }
+ }
+ Mask = DAG.getNode(ISD::BUILD_VECTOR, getCurDebugLoc(),
+ Mask.getValueType(),
+ &MappedOps[0], MappedOps.size());
+ setValue(&I, DAG.getNode(ISD::VECTOR_SHUFFLE, getCurDebugLoc(),
+ VT, Src1, Src2, Mask));
+ return;
+ }
+ }
+
+ // We can't use either concat vectors or extract subvectors so fall back to
+ // replacing the shuffle with extract and build vector.
+ // to insert and build vector.
+ MVT EltVT = VT.getVectorElementType();
+ MVT PtrVT = TLI.getPointerTy();
+ SmallVector<SDValue,8> Ops;
+ for (int i = 0; i != MaskNumElts; ++i) {
+ SDValue Arg = Mask.getOperand(i);
+ if (Arg.getOpcode() == ISD::UNDEF) {
+ Ops.push_back(DAG.getNode(ISD::UNDEF, getCurDebugLoc(), EltVT));
+ } else {
+ assert(isa<ConstantSDNode>(Arg) && "Invalid VECTOR_SHUFFLE mask!");
+ int Idx = cast<ConstantSDNode>(Arg)->getZExtValue();
+ if (Idx < SrcNumElts)
+ Ops.push_back(DAG.getNode(ISD::EXTRACT_VECTOR_ELT, getCurDebugLoc(),
+ EltVT, Src1, DAG.getConstant(Idx, PtrVT)));
+ else
+ Ops.push_back(DAG.getNode(ISD::EXTRACT_VECTOR_ELT, getCurDebugLoc(),
+ EltVT, Src2,
+ DAG.getConstant(Idx - SrcNumElts, PtrVT)));
+ }
+ }
+ setValue(&I, DAG.getNode(ISD::BUILD_VECTOR, getCurDebugLoc(),
+ VT, &Ops[0], Ops.size()));
}
void SelectionDAGLowering::visitInsertValue(InsertValueInst &I) {
unsigned i = 0;
// Copy the beginning value(s) from the original aggregate.
for (; i != LinearIndex; ++i)
- Values[i] = IntoUndef ? DAG.getNode(ISD::UNDEF, AggValueVTs[i]) :
+ Values[i] = IntoUndef ? DAG.getNode(ISD::UNDEF, getCurDebugLoc(),
+ AggValueVTs[i]) :
SDValue(Agg.getNode(), Agg.getResNo() + i);
// Copy values from the inserted value(s).
for (; i != LinearIndex + NumValValues; ++i)
- Values[i] = FromUndef ? DAG.getNode(ISD::UNDEF, AggValueVTs[i]) :
+ Values[i] = FromUndef ? DAG.getNode(ISD::UNDEF, getCurDebugLoc(),
+ AggValueVTs[i]) :
SDValue(Val.getNode(), Val.getResNo() + i - LinearIndex);
// Copy remaining value(s) from the original aggregate.
for (; i != NumAggValues; ++i)
- Values[i] = IntoUndef ? DAG.getNode(ISD::UNDEF, AggValueVTs[i]) :
+ Values[i] = IntoUndef ? DAG.getNode(ISD::UNDEF, getCurDebugLoc(),
+ AggValueVTs[i]) :
SDValue(Agg.getNode(), Agg.getResNo() + i);
- setValue(&I, DAG.getMergeValues(DAG.getVTList(&AggValueVTs[0], NumAggValues),
- &Values[0], NumAggValues));
+ setValue(&I, DAG.getNode(ISD::MERGE_VALUES, getCurDebugLoc(),
+ DAG.getVTList(&AggValueVTs[0], NumAggValues),
+ &Values[0], NumAggValues));
}
void SelectionDAGLowering::visitExtractValue(ExtractValueInst &I) {
// Copy out the selected value(s).
for (unsigned i = LinearIndex; i != LinearIndex + NumValValues; ++i)
Values[i - LinearIndex] =
- OutOfUndef ? DAG.getNode(ISD::UNDEF, Agg.getNode()->getValueType(Agg.getResNo() + i)) :
- SDValue(Agg.getNode(), Agg.getResNo() + i);
-
- setValue(&I, DAG.getMergeValues(DAG.getVTList(&ValValueVTs[0], NumValValues),
- &Values[0], NumValValues));
+ OutOfUndef ?
+ DAG.getNode(ISD::UNDEF, getCurDebugLoc(),
+ Agg.getNode()->getValueType(Agg.getResNo() + i)) :
+ SDValue(Agg.getNode(), Agg.getResNo() + i);
+
+ setValue(&I, DAG.getNode(ISD::MERGE_VALUES, getCurDebugLoc(),
+ DAG.getVTList(&ValValueVTs[0], NumValValues),
+ &Values[0], NumValValues));
}
if (Field) {
// N = N + Offset
uint64_t Offset = TD->getStructLayout(StTy)->getElementOffset(Field);
- N = DAG.getNode(ISD::ADD, N.getValueType(), N,
+ N = DAG.getNode(ISD::ADD, getCurDebugLoc(), N.getValueType(), N,
DAG.getIntPtrConstant(Offset));
}
Ty = StTy->getElementType(Field);
// If this is a constant subscript, handle it quickly.
if (ConstantInt *CI = dyn_cast<ConstantInt>(Idx)) {
if (CI->getZExtValue() == 0) continue;
- uint64_t Offs =
- TD->getABITypeSize(Ty)*cast<ConstantInt>(CI)->getSExtValue();
- N = DAG.getNode(ISD::ADD, N.getValueType(), N,
+ uint64_t Offs =
+ TD->getTypePaddedSize(Ty)*cast<ConstantInt>(CI)->getSExtValue();
+ N = DAG.getNode(ISD::ADD, getCurDebugLoc(), N.getValueType(), N,
DAG.getIntPtrConstant(Offs));
continue;
}
-
+
// N = N + Idx * ElementSize;
- uint64_t ElementSize = TD->getABITypeSize(Ty);
+ uint64_t ElementSize = TD->getTypePaddedSize(Ty);
SDValue IdxN = getValue(Idx);
// If the index is smaller or larger than intptr_t, truncate or extend
// it.
if (IdxN.getValueType().bitsLT(N.getValueType()))
- IdxN = DAG.getNode(ISD::SIGN_EXTEND, N.getValueType(), IdxN);
+ IdxN = DAG.getNode(ISD::SIGN_EXTEND, getCurDebugLoc(),
+ N.getValueType(), IdxN);
else if (IdxN.getValueType().bitsGT(N.getValueType()))
- IdxN = DAG.getNode(ISD::TRUNCATE, N.getValueType(), IdxN);
+ IdxN = DAG.getNode(ISD::TRUNCATE, getCurDebugLoc(),
+ N.getValueType(), IdxN);
// If this is a multiply by a power of two, turn it into a shl
// immediately. This is a very common case.
if (ElementSize != 1) {
if (isPowerOf2_64(ElementSize)) {
unsigned Amt = Log2_64(ElementSize);
- IdxN = DAG.getNode(ISD::SHL, N.getValueType(), IdxN,
- DAG.getConstant(Amt, TLI.getShiftAmountTy()));
+ IdxN = DAG.getNode(ISD::SHL, getCurDebugLoc(),
+ N.getValueType(), IdxN,
+ DAG.getConstant(Amt, TLI.getPointerTy()));
} else {
SDValue Scale = DAG.getIntPtrConstant(ElementSize);
- IdxN = DAG.getNode(ISD::MUL, N.getValueType(), IdxN, Scale);
+ IdxN = DAG.getNode(ISD::MUL, getCurDebugLoc(),
+ N.getValueType(), IdxN, Scale);
}
}
- N = DAG.getNode(ISD::ADD, N.getValueType(), N, IdxN);
+ N = DAG.getNode(ISD::ADD, getCurDebugLoc(),
+ N.getValueType(), N, IdxN);
}
}
setValue(&I, N);
return; // getValue will auto-populate this.
const Type *Ty = I.getAllocatedType();
- uint64_t TySize = TLI.getTargetData()->getABITypeSize(Ty);
+ uint64_t TySize = TLI.getTargetData()->getTypePaddedSize(Ty);
unsigned Align =
std::max((unsigned)TLI.getTargetData()->getPrefTypeAlignment(Ty),
I.getAlignment());
SDValue AllocSize = getValue(I.getArraySize());
MVT IntPtr = TLI.getPointerTy();
if (IntPtr.bitsLT(AllocSize.getValueType()))
- AllocSize = DAG.getNode(ISD::TRUNCATE, IntPtr, AllocSize);
+ AllocSize = DAG.getNode(ISD::TRUNCATE, getCurDebugLoc(),
+ IntPtr, AllocSize);
else if (IntPtr.bitsGT(AllocSize.getValueType()))
- AllocSize = DAG.getNode(ISD::ZERO_EXTEND, IntPtr, AllocSize);
+ AllocSize = DAG.getNode(ISD::ZERO_EXTEND, getCurDebugLoc(),
+ IntPtr, AllocSize);
- AllocSize = DAG.getNode(ISD::MUL, IntPtr, AllocSize,
+ AllocSize = DAG.getNode(ISD::MUL, getCurDebugLoc(), IntPtr, AllocSize,
DAG.getIntPtrConstant(TySize));
// Handle alignment. If the requested alignment is less than or equal to
// Round the size of the allocation up to the stack alignment size
// by add SA-1 to the size.
- AllocSize = DAG.getNode(ISD::ADD, AllocSize.getValueType(), AllocSize,
+ AllocSize = DAG.getNode(ISD::ADD, getCurDebugLoc(),
+ AllocSize.getValueType(), AllocSize,
DAG.getIntPtrConstant(StackAlign-1));
// Mask out the low bits for alignment purposes.
- AllocSize = DAG.getNode(ISD::AND, AllocSize.getValueType(), AllocSize,
+ AllocSize = DAG.getNode(ISD::AND, getCurDebugLoc(),
+ AllocSize.getValueType(), AllocSize,
DAG.getIntPtrConstant(~(uint64_t)(StackAlign-1)));
SDValue Ops[] = { getRoot(), AllocSize, DAG.getIntPtrConstant(Align) };
const MVT *VTs = DAG.getNodeValueTypes(AllocSize.getValueType(),
MVT::Other);
- SDValue DSA = DAG.getNode(ISD::DYNAMIC_STACKALLOC, VTs, 2, Ops, 3);
+ SDValue DSA = DAG.getNode(ISD::DYNAMIC_STACKALLOC, getCurDebugLoc(),
+ VTs, 2, Ops, 3);
setValue(&I, DSA);
DAG.setRoot(DSA.getValue(1));
SmallVector<SDValue, 4> Chains(NumValues);
MVT PtrVT = Ptr.getValueType();
for (unsigned i = 0; i != NumValues; ++i) {
- SDValue L = DAG.getLoad(ValueVTs[i], Root,
- DAG.getNode(ISD::ADD, PtrVT, Ptr,
+ SDValue L = DAG.getLoad(ValueVTs[i], getCurDebugLoc(), Root,
+ DAG.getNode(ISD::ADD, getCurDebugLoc(),
+ PtrVT, Ptr,
DAG.getConstant(Offsets[i], PtrVT)),
SV, Offsets[i],
isVolatile, Alignment);
Values[i] = L;
Chains[i] = L.getValue(1);
}
-
+
if (!ConstantMemory) {
- SDValue Chain = DAG.getNode(ISD::TokenFactor, MVT::Other,
+ SDValue Chain = DAG.getNode(ISD::TokenFactor, getCurDebugLoc(),
+ MVT::Other,
&Chains[0], NumValues);
if (isVolatile)
DAG.setRoot(Chain);
PendingLoads.push_back(Chain);
}
- setValue(&I, DAG.getMergeValues(DAG.getVTList(&ValueVTs[0], NumValues),
- &Values[0], NumValues));
+ setValue(&I, DAG.getNode(ISD::MERGE_VALUES, getCurDebugLoc(),
+ DAG.getVTList(&ValueVTs[0], NumValues),
+ &Values[0], NumValues));
}
bool isVolatile = I.isVolatile();
unsigned Alignment = I.getAlignment();
for (unsigned i = 0; i != NumValues; ++i)
- Chains[i] = DAG.getStore(Root, SDValue(Src.getNode(), Src.getResNo() + i),
- DAG.getNode(ISD::ADD, PtrVT, Ptr,
+ Chains[i] = DAG.getStore(Root, getCurDebugLoc(),
+ SDValue(Src.getNode(), Src.getResNo() + i),
+ DAG.getNode(ISD::ADD, getCurDebugLoc(),
+ PtrVT, Ptr,
DAG.getConstant(Offsets[i], PtrVT)),
PtrV, Offsets[i],
isVolatile, Alignment);
- DAG.setRoot(DAG.getNode(ISD::TokenFactor, MVT::Other, &Chains[0], NumValues));
+ DAG.setRoot(DAG.getNode(ISD::TokenFactor, getCurDebugLoc(),
+ MVT::Other, &Chains[0], NumValues));
}
/// visitTargetIntrinsic - Lower a call of a target intrinsic to an INTRINSIC
/// node.
-void SelectionDAGLowering::visitTargetIntrinsic(CallInst &I,
+void SelectionDAGLowering::visitTargetIntrinsic(CallInst &I,
unsigned Intrinsic) {
bool HasChain = !I.doesNotAccessMemory();
bool OnlyLoad = HasChain && I.onlyReadsMemory();
if (OnlyLoad) {
// We don't need to serialize loads against other loads.
Ops.push_back(DAG.getRoot());
- } else {
+ } else {
Ops.push_back(getRoot());
}
}
-
- // Add the intrinsic ID as an integer operand.
- Ops.push_back(DAG.getConstant(Intrinsic, TLI.getPointerTy()));
+
+ // Info is set by getTgtMemInstrinsic
+ TargetLowering::IntrinsicInfo Info;
+ bool IsTgtIntrinsic = TLI.getTgtMemIntrinsic(Info, I, Intrinsic);
+
+ // Add the intrinsic ID as an integer operand if it's not a target intrinsic.
+ if (!IsTgtIntrinsic)
+ Ops.push_back(DAG.getConstant(Intrinsic, TLI.getPointerTy()));
// Add all operands of the call to the operand list.
for (unsigned i = 1, e = I.getNumOperands(); i != e; ++i) {
if (VT.isVector()) {
const VectorType *DestTy = cast<VectorType>(I.getType());
MVT EltVT = TLI.getValueType(DestTy->getElementType());
-
+
VT = MVT::getVectorVT(EltVT, DestTy->getNumElements());
assert(VT != MVT::Other && "Intrinsic uses a non-legal type?");
}
-
+
assert(TLI.isTypeLegal(VT) && "Intrinsic uses a non-legal type?");
VTs.push_back(VT);
}
// Create the node.
SDValue Result;
- if (!HasChain)
- Result = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, VTList, VTs.size(),
+ if (IsTgtIntrinsic) {
+ // This is target intrinsic that touches memory
+ Result = DAG.getMemIntrinsicNode(Info.opc, getCurDebugLoc(),
+ VTList, VTs.size(),
+ &Ops[0], Ops.size(),
+ Info.memVT, Info.ptrVal, Info.offset,
+ Info.align, Info.vol,
+ Info.readMem, Info.writeMem);
+ }
+ else if (!HasChain)
+ Result = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, getCurDebugLoc(),
+ VTList, VTs.size(),
&Ops[0], Ops.size());
else if (I.getType() != Type::VoidTy)
- Result = DAG.getNode(ISD::INTRINSIC_W_CHAIN, VTList, VTs.size(),
+ Result = DAG.getNode(ISD::INTRINSIC_W_CHAIN, getCurDebugLoc(),
+ VTList, VTs.size(),
&Ops[0], Ops.size());
else
- Result = DAG.getNode(ISD::INTRINSIC_VOID, VTList, VTs.size(),
+ Result = DAG.getNode(ISD::INTRINSIC_VOID, getCurDebugLoc(),
+ VTList, VTs.size(),
&Ops[0], Ops.size());
if (HasChain) {
if (I.getType() != Type::VoidTy) {
if (const VectorType *PTy = dyn_cast<VectorType>(I.getType())) {
MVT VT = TLI.getValueType(PTy);
- Result = DAG.getNode(ISD::BIT_CONVERT, VT, Result);
- }
+ Result = DAG.getNode(ISD::BIT_CONVERT, getCurDebugLoc(), VT, Result);
+ }
setValue(&I, Result);
}
}
///
/// where Op is the hexidecimal representation of floating point value.
static SDValue
-GetSignificand(SelectionDAG &DAG, SDValue Op) {
- SDValue t1 = DAG.getNode(ISD::AND, MVT::i32, Op,
- DAG.getConstant(0x007fffff, MVT::i32));
- SDValue t2 = DAG.getNode(ISD::OR, MVT::i32, t1,
- DAG.getConstant(0x3f800000, MVT::i32));
- return DAG.getNode(ISD::BIT_CONVERT, MVT::f32, t2);
+GetSignificand(SelectionDAG &DAG, SDValue Op, DebugLoc dl) {
+ SDValue t1 = DAG.getNode(ISD::AND, dl, MVT::i32, Op,
+ DAG.getConstant(0x007fffff, MVT::i32));
+ SDValue t2 = DAG.getNode(ISD::OR, dl, MVT::i32, t1,
+ DAG.getConstant(0x3f800000, MVT::i32));
+ return DAG.getNode(ISD::BIT_CONVERT, dl, MVT::f32, t2);
}
/// GetExponent - Get the exponent:
///
-/// (float)((Op1 >> 23) - 127);
+/// (float)(int)(((Op & 0x7f800000) >> 23) - 127);
///
/// where Op is the hexidecimal representation of floating point value.
static SDValue
-GetExponent(SelectionDAG &DAG, SDValue Op) {
- SDValue t1 = DAG.getNode(ISD::SRL, MVT::i32, Op,
- DAG.getConstant(23, MVT::i32));
- SDValue t2 = DAG.getNode(ISD::SUB, MVT::i32, t1,
- DAG.getConstant(127, MVT::i32));
- return DAG.getNode(ISD::UINT_TO_FP, MVT::f32, t2);
+GetExponent(SelectionDAG &DAG, SDValue Op, const TargetLowering &TLI,
+ DebugLoc dl) {
+ SDValue t0 = DAG.getNode(ISD::AND, dl, MVT::i32, Op,
+ DAG.getConstant(0x7f800000, MVT::i32));
+ SDValue t1 = DAG.getNode(ISD::SRL, dl, MVT::i32, t0,
+ DAG.getConstant(23, TLI.getPointerTy()));
+ SDValue t2 = DAG.getNode(ISD::SUB, dl, MVT::i32, t1,
+ DAG.getConstant(127, MVT::i32));
+ return DAG.getNode(ISD::SINT_TO_FP, dl, MVT::f32, t2);
}
/// getF32Constant - Get 32-bit floating point constant.
return DAG.getConstantFP(APFloat(APInt(32, Flt)), MVT::f32);
}
-/// Inlined utility function to implement binary input atomic intrinsics for
+/// Inlined utility function to implement binary input atomic intrinsics for
/// visitIntrinsicCall: I is a call instruction
/// Op is the associated NodeType for I
const char *
SelectionDAGLowering::implVisitBinaryAtomic(CallInst& I, ISD::NodeType Op) {
- SDValue Root = getRoot();
- SDValue L = DAG.getAtomic(Op, Root,
- getValue(I.getOperand(1)),
- getValue(I.getOperand(2)),
- I.getOperand(1));
+ SDValue Root = getRoot();
+ SDValue L =
+ DAG.getAtomic(Op, getCurDebugLoc(),
+ getValue(I.getOperand(2)).getValueType().getSimpleVT(),
+ Root,
+ getValue(I.getOperand(1)),
+ getValue(I.getOperand(2)),
+ I.getOperand(1));
setValue(&I, L);
DAG.setRoot(L.getValue(1));
return 0;
}
+// implVisitAluOverflow - Lower arithmetic overflow instrinsics.
+const char *
+SelectionDAGLowering::implVisitAluOverflow(CallInst &I, ISD::NodeType Op) {
+ SDValue Op1 = getValue(I.getOperand(1));
+ SDValue Op2 = getValue(I.getOperand(2));
+
+ MVT ValueVTs[] = { Op1.getValueType(), MVT::i1 };
+ SDValue Ops[] = { Op1, Op2 };
+
+ SDValue Result = DAG.getNode(Op, getCurDebugLoc(),
+ DAG.getVTList(&ValueVTs[0], 2), &Ops[0], 2);
+
+ setValue(&I, Result);
+ return 0;
+}
+
/// visitExp - Lower an exp intrinsic. Handles the special sequences for
/// limited-precision mode.
void
SelectionDAGLowering::visitExp(CallInst &I) {
SDValue result;
+ DebugLoc dl = getCurDebugLoc();
if (getValue(I.getOperand(1)).getValueType() == MVT::f32 &&
LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
//
// #define LOG2OFe 1.4426950f
// IntegerPartOfX = ((int32_t)(X * LOG2OFe));
- SDValue t0 = DAG.getNode(ISD::FMUL, MVT::f32, Op,
+ SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, Op,
getF32Constant(DAG, 0x3fb8aa3b));
- SDValue IntegerPartOfX = DAG.getNode(ISD::FP_TO_SINT, MVT::i32, t0);
+ SDValue IntegerPartOfX = DAG.getNode(ISD::FP_TO_SINT, dl, MVT::i32, t0);
// FractionalPartOfX = (X * LOG2OFe) - (float)IntegerPartOfX;
- SDValue t1 = DAG.getNode(ISD::SINT_TO_FP, MVT::f32, IntegerPartOfX);
- SDValue X = DAG.getNode(ISD::FSUB, MVT::f32, t0, t1);
+ SDValue t1 = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::f32, IntegerPartOfX);
+ SDValue X = DAG.getNode(ISD::FSUB, dl, MVT::f32, t0, t1);
// IntegerPartOfX <<= 23;
- IntegerPartOfX = DAG.getNode(ISD::SHL, MVT::i32, IntegerPartOfX,
- DAG.getConstant(23, MVT::i32));
+ IntegerPartOfX = DAG.getNode(ISD::SHL, dl, MVT::i32, IntegerPartOfX,
+ DAG.getConstant(23, TLI.getPointerTy()));
if (LimitFloatPrecision <= 6) {
// For floating-point precision of 6:
// (0.735607626f + 0.252464424f * x) * x;
//
// error 0.0144103317, which is 6 bits
- SDValue t2 = DAG.getNode(ISD::FMUL, MVT::f32, X,
+ SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0x3e814304));
- SDValue t3 = DAG.getNode(ISD::FADD, MVT::f32, t2,
+ SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3f3c50c8));
- SDValue t4 = DAG.getNode(ISD::FMUL, MVT::f32, t3, X);
- SDValue t5 = DAG.getNode(ISD::FADD, MVT::f32, t4,
+ SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
+ SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x3f7f5e7e));
- SDValue TwoToFracPartOfX = DAG.getNode(ISD::BIT_CONVERT, MVT::i32, t5);
+ SDValue TwoToFracPartOfX = DAG.getNode(ISD::BIT_CONVERT, dl,MVT::i32, t5);
// Add the exponent into the result in integer domain.
- SDValue t6 = DAG.getNode(ISD::ADD, MVT::i32,
+ SDValue t6 = DAG.getNode(ISD::ADD, dl, MVT::i32,
TwoToFracPartOfX, IntegerPartOfX);
- result = DAG.getNode(ISD::BIT_CONVERT, MVT::f32, t6);
+ result = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::f32, t6);
} else if (LimitFloatPrecision > 6 && LimitFloatPrecision <= 12) {
// For floating-point precision of 12:
//
// (0.224338339f + 0.792043434e-1f * x) * x) * x;
//
// 0.000107046256 error, which is 13 to 14 bits
- SDValue t2 = DAG.getNode(ISD::FMUL, MVT::f32, X,
+ SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0x3da235e3));
- SDValue t3 = DAG.getNode(ISD::FADD, MVT::f32, t2,
+ SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3e65b8f3));
- SDValue t4 = DAG.getNode(ISD::FMUL, MVT::f32, t3, X);
- SDValue t5 = DAG.getNode(ISD::FADD, MVT::f32, t4,
+ SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
+ SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x3f324b07));
- SDValue t6 = DAG.getNode(ISD::FMUL, MVT::f32, t5, X);
- SDValue t7 = DAG.getNode(ISD::FADD, MVT::f32, t6,
+ SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
+ SDValue t7 = DAG.getNode(ISD::FADD, dl, MVT::f32, t6,
getF32Constant(DAG, 0x3f7ff8fd));
- SDValue TwoToFracPartOfX = DAG.getNode(ISD::BIT_CONVERT, MVT::i32, t7);
+ SDValue TwoToFracPartOfX = DAG.getNode(ISD::BIT_CONVERT, dl,MVT::i32, t7);
// Add the exponent into the result in integer domain.
- SDValue t8 = DAG.getNode(ISD::ADD, MVT::i32,
+ SDValue t8 = DAG.getNode(ISD::ADD, dl, MVT::i32,
TwoToFracPartOfX, IntegerPartOfX);
- result = DAG.getNode(ISD::BIT_CONVERT, MVT::f32, t8);
+ result = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::f32, t8);
} else { // LimitFloatPrecision > 12 && LimitFloatPrecision <= 18
// For floating-point precision of 18:
//
// (0.136028312e-2f + 0.157059148e-3f *x)*x)*x)*x)*x)*x;
//
// error 2.47208000*10^(-7), which is better than 18 bits
- SDValue t2 = DAG.getNode(ISD::FMUL, MVT::f32, X,
+ SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0x3924b03e));
- SDValue t3 = DAG.getNode(ISD::FADD, MVT::f32, t2,
+ SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3ab24b87));
- SDValue t4 = DAG.getNode(ISD::FMUL, MVT::f32, t3, X);
- SDValue t5 = DAG.getNode(ISD::FADD, MVT::f32, t4,
+ SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
+ SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x3c1d8c17));
- SDValue t6 = DAG.getNode(ISD::FMUL, MVT::f32, t5, X);
- SDValue t7 = DAG.getNode(ISD::FADD, MVT::f32, t6,
+ SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
+ SDValue t7 = DAG.getNode(ISD::FADD, dl, MVT::f32, t6,
getF32Constant(DAG, 0x3d634a1d));
- SDValue t8 = DAG.getNode(ISD::FMUL, MVT::f32, t7, X);
- SDValue t9 = DAG.getNode(ISD::FADD, MVT::f32, t8,
+ SDValue t8 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t7, X);
+ SDValue t9 = DAG.getNode(ISD::FADD, dl, MVT::f32, t8,
getF32Constant(DAG, 0x3e75fe14));
- SDValue t10 = DAG.getNode(ISD::FMUL, MVT::f32, t9, X);
- SDValue t11 = DAG.getNode(ISD::FADD, MVT::f32, t10,
+ SDValue t10 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t9, X);
+ SDValue t11 = DAG.getNode(ISD::FADD, dl, MVT::f32, t10,
getF32Constant(DAG, 0x3f317234));
- SDValue t12 = DAG.getNode(ISD::FMUL, MVT::f32, t11, X);
- SDValue t13 = DAG.getNode(ISD::FADD, MVT::f32, t12,
+ SDValue t12 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t11, X);
+ SDValue t13 = DAG.getNode(ISD::FADD, dl, MVT::f32, t12,
getF32Constant(DAG, 0x3f800000));
- SDValue TwoToFracPartOfX = DAG.getNode(ISD::BIT_CONVERT, MVT::i32, t13);
+ SDValue TwoToFracPartOfX = DAG.getNode(ISD::BIT_CONVERT, dl,
+ MVT::i32, t13);
// Add the exponent into the result in integer domain.
- SDValue t14 = DAG.getNode(ISD::ADD, MVT::i32,
+ SDValue t14 = DAG.getNode(ISD::ADD, dl, MVT::i32,
TwoToFracPartOfX, IntegerPartOfX);
- result = DAG.getNode(ISD::BIT_CONVERT, MVT::f32, t14);
+ result = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::f32, t14);
}
} else {
// No special expansion.
- result = DAG.getNode(ISD::FEXP,
+ result = DAG.getNode(ISD::FEXP, dl,
getValue(I.getOperand(1)).getValueType(),
getValue(I.getOperand(1)));
}
void
SelectionDAGLowering::visitLog(CallInst &I) {
SDValue result;
+ DebugLoc dl = getCurDebugLoc();
if (getValue(I.getOperand(1)).getValueType() == MVT::f32 &&
LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
SDValue Op = getValue(I.getOperand(1));
- SDValue Op1 = DAG.getNode(ISD::BIT_CONVERT, MVT::i32, Op);
+ SDValue Op1 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, Op);
// Scale the exponent by log(2) [0.69314718f].
- SDValue Exp = GetExponent(DAG, Op1);
- SDValue LogOfExponent = DAG.getNode(ISD::FMUL, MVT::f32, Exp,
+ SDValue Exp = GetExponent(DAG, Op1, TLI, dl);
+ SDValue LogOfExponent = DAG.getNode(ISD::FMUL, dl, MVT::f32, Exp,
getF32Constant(DAG, 0x3f317218));
// Get the significand and build it into a floating-point number with
// exponent of 1.
- SDValue X = GetSignificand(DAG, Op1);
+ SDValue X = GetSignificand(DAG, Op1, dl);
if (LimitFloatPrecision <= 6) {
// For floating-point precision of 6:
// LogofMantissa =
// -1.1609546f +
// (1.4034025f - 0.23903021f * x) * x;
- //
+ //
// error 0.0034276066, which is better than 8 bits
- SDValue t0 = DAG.getNode(ISD::FMUL, MVT::f32, X,
+ SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0xbe74c456));
- SDValue t1 = DAG.getNode(ISD::FADD, MVT::f32, t0,
+ SDValue t1 = DAG.getNode(ISD::FADD, dl, MVT::f32, t0,
getF32Constant(DAG, 0x3fb3a2b1));
- SDValue t2 = DAG.getNode(ISD::FMUL, MVT::f32, t1, X);
- SDValue LogOfMantissa = DAG.getNode(ISD::FSUB, MVT::f32, t2,
+ SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
+ SDValue LogOfMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3f949a29));
- result = DAG.getNode(ISD::FADD, MVT::f32, LogOfExponent, LogOfMantissa);
+ result = DAG.getNode(ISD::FADD, dl,
+ MVT::f32, LogOfExponent, LogOfMantissa);
} else if (LimitFloatPrecision > 6 && LimitFloatPrecision <= 12) {
// For floating-point precision of 12:
//
// (0.44717955f - 0.56570851e-1f * x) * x) * x) * x;
//
// error 0.000061011436, which is 14 bits
- SDValue t0 = DAG.getNode(ISD::FMUL, MVT::f32, X,
+ SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0xbd67b6d6));
- SDValue t1 = DAG.getNode(ISD::FADD, MVT::f32, t0,
+ SDValue t1 = DAG.getNode(ISD::FADD, dl, MVT::f32, t0,
getF32Constant(DAG, 0x3ee4f4b8));
- SDValue t2 = DAG.getNode(ISD::FMUL, MVT::f32, t1, X);
- SDValue t3 = DAG.getNode(ISD::FSUB, MVT::f32, t2,
+ SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
+ SDValue t3 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3fbc278b));
- SDValue t4 = DAG.getNode(ISD::FMUL, MVT::f32, t3, X);
- SDValue t5 = DAG.getNode(ISD::FADD, MVT::f32, t4,
+ SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
+ SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x40348e95));
- SDValue t6 = DAG.getNode(ISD::FMUL, MVT::f32, t5, X);
- SDValue LogOfMantissa = DAG.getNode(ISD::FSUB, MVT::f32, t6,
+ SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
+ SDValue LogOfMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t6,
getF32Constant(DAG, 0x3fdef31a));
- result = DAG.getNode(ISD::FADD, MVT::f32, LogOfExponent, LogOfMantissa);
+ result = DAG.getNode(ISD::FADD, dl,
+ MVT::f32, LogOfExponent, LogOfMantissa);
} else { // LimitFloatPrecision > 12 && LimitFloatPrecision <= 18
// For floating-point precision of 18:
//
// (0.19073739f - 0.17809712e-1f * x) * x) * x) * x) * x)*x;
//
// error 0.0000023660568, which is better than 18 bits
- SDValue t0 = DAG.getNode(ISD::FMUL, MVT::f32, X,
+ SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0xbc91e5ac));
- SDValue t1 = DAG.getNode(ISD::FADD, MVT::f32, t0,
+ SDValue t1 = DAG.getNode(ISD::FADD, dl, MVT::f32, t0,
getF32Constant(DAG, 0x3e4350aa));
- SDValue t2 = DAG.getNode(ISD::FMUL, MVT::f32, t1, X);
- SDValue t3 = DAG.getNode(ISD::FSUB, MVT::f32, t2,
+ SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
+ SDValue t3 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3f60d3e3));
- SDValue t4 = DAG.getNode(ISD::FMUL, MVT::f32, t3, X);
- SDValue t5 = DAG.getNode(ISD::FADD, MVT::f32, t4,
+ SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
+ SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x4011cdf0));
- SDValue t6 = DAG.getNode(ISD::FMUL, MVT::f32, t5, X);
- SDValue t7 = DAG.getNode(ISD::FSUB, MVT::f32, t6,
+ SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
+ SDValue t7 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t6,
getF32Constant(DAG, 0x406cfd1c));
- SDValue t8 = DAG.getNode(ISD::FMUL, MVT::f32, t7, X);
- SDValue t9 = DAG.getNode(ISD::FADD, MVT::f32, t8,
+ SDValue t8 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t7, X);
+ SDValue t9 = DAG.getNode(ISD::FADD, dl, MVT::f32, t8,
getF32Constant(DAG, 0x408797cb));
- SDValue t10 = DAG.getNode(ISD::FMUL, MVT::f32, t9, X);
- SDValue LogOfMantissa = DAG.getNode(ISD::FSUB, MVT::f32, t10,
+ SDValue t10 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t9, X);
+ SDValue LogOfMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t10,
getF32Constant(DAG, 0x4006dcab));
- result = DAG.getNode(ISD::FADD, MVT::f32, LogOfExponent, LogOfMantissa);
+ result = DAG.getNode(ISD::FADD, dl,
+ MVT::f32, LogOfExponent, LogOfMantissa);
}
} else {
// No special expansion.
- result = DAG.getNode(ISD::FLOG,
+ result = DAG.getNode(ISD::FLOG, dl,
getValue(I.getOperand(1)).getValueType(),
getValue(I.getOperand(1)));
}
void
SelectionDAGLowering::visitLog2(CallInst &I) {
SDValue result;
+ DebugLoc dl = getCurDebugLoc();
if (getValue(I.getOperand(1)).getValueType() == MVT::f32 &&
LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
SDValue Op = getValue(I.getOperand(1));
- SDValue Op1 = DAG.getNode(ISD::BIT_CONVERT, MVT::i32, Op);
+ SDValue Op1 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, Op);
// Get the exponent.
- SDValue LogOfExponent = GetExponent(DAG, Op1);
+ SDValue LogOfExponent = GetExponent(DAG, Op1, TLI, dl);
// Get the significand and build it into a floating-point number with
// exponent of 1.
- SDValue X = GetSignificand(DAG, Op1);
-
+ SDValue X = GetSignificand(DAG, Op1, dl);
+
// Different possible minimax approximations of significand in
// floating-point for various degrees of accuracy over [1,2].
if (LimitFloatPrecision <= 6) {
// Log2ofMantissa = -1.6749035f + (2.0246817f - .34484768f * x) * x;
//
// error 0.0049451742, which is more than 7 bits
- SDValue t0 = DAG.getNode(ISD::FMUL, MVT::f32, X,
+ SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0xbeb08fe0));
- SDValue t1 = DAG.getNode(ISD::FADD, MVT::f32, t0,
+ SDValue t1 = DAG.getNode(ISD::FADD, dl, MVT::f32, t0,
getF32Constant(DAG, 0x40019463));
- SDValue t2 = DAG.getNode(ISD::FMUL, MVT::f32, t1, X);
- SDValue Log2ofMantissa = DAG.getNode(ISD::FSUB, MVT::f32, t2,
+ SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
+ SDValue Log2ofMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3fd6633d));
- result = DAG.getNode(ISD::FADD, MVT::f32, LogOfExponent, Log2ofMantissa);
+ result = DAG.getNode(ISD::FADD, dl,
+ MVT::f32, LogOfExponent, Log2ofMantissa);
} else if (LimitFloatPrecision > 6 && LimitFloatPrecision <= 12) {
// For floating-point precision of 12:
//
// (4.07009056f +
// (-2.12067489f +
// (.645142248f - 0.816157886e-1f * x) * x) * x) * x;
- //
+ //
// error 0.0000876136000, which is better than 13 bits
- SDValue t0 = DAG.getNode(ISD::FMUL, MVT::f32, X,
+ SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0xbda7262e));
- SDValue t1 = DAG.getNode(ISD::FADD, MVT::f32, t0,
+ SDValue t1 = DAG.getNode(ISD::FADD, dl, MVT::f32, t0,
getF32Constant(DAG, 0x3f25280b));
- SDValue t2 = DAG.getNode(ISD::FMUL, MVT::f32, t1, X);
- SDValue t3 = DAG.getNode(ISD::FSUB, MVT::f32, t2,
+ SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
+ SDValue t3 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t2,
getF32Constant(DAG, 0x4007b923));
- SDValue t4 = DAG.getNode(ISD::FMUL, MVT::f32, t3, X);
- SDValue t5 = DAG.getNode(ISD::FADD, MVT::f32, t4,
+ SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
+ SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x40823e2f));
- SDValue t6 = DAG.getNode(ISD::FMUL, MVT::f32, t5, X);
- SDValue Log2ofMantissa = DAG.getNode(ISD::FSUB, MVT::f32, t6,
+ SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
+ SDValue Log2ofMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t6,
getF32Constant(DAG, 0x4020d29c));
- result = DAG.getNode(ISD::FADD, MVT::f32, LogOfExponent, Log2ofMantissa);
+ result = DAG.getNode(ISD::FADD, dl,
+ MVT::f32, LogOfExponent, Log2ofMantissa);
} else { // LimitFloatPrecision > 12 && LimitFloatPrecision <= 18
// For floating-point precision of 18:
//
// 0.25691327e-1f * x) * x) * x) * x) * x) * x;
//
// error 0.0000018516, which is better than 18 bits
- SDValue t0 = DAG.getNode(ISD::FMUL, MVT::f32, X,
+ SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0xbcd2769e));
- SDValue t1 = DAG.getNode(ISD::FADD, MVT::f32, t0,
+ SDValue t1 = DAG.getNode(ISD::FADD, dl, MVT::f32, t0,
getF32Constant(DAG, 0x3e8ce0b9));
- SDValue t2 = DAG.getNode(ISD::FMUL, MVT::f32, t1, X);
- SDValue t3 = DAG.getNode(ISD::FSUB, MVT::f32, t2,
+ SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
+ SDValue t3 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3fa22ae7));
- SDValue t4 = DAG.getNode(ISD::FMUL, MVT::f32, t3, X);
- SDValue t5 = DAG.getNode(ISD::FADD, MVT::f32, t4,
+ SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
+ SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x40525723));
- SDValue t6 = DAG.getNode(ISD::FMUL, MVT::f32, t5, X);
- SDValue t7 = DAG.getNode(ISD::FSUB, MVT::f32, t6,
+ SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
+ SDValue t7 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t6,
getF32Constant(DAG, 0x40aaf200));
- SDValue t8 = DAG.getNode(ISD::FMUL, MVT::f32, t7, X);
- SDValue t9 = DAG.getNode(ISD::FADD, MVT::f32, t8,
+ SDValue t8 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t7, X);
+ SDValue t9 = DAG.getNode(ISD::FADD, dl, MVT::f32, t8,
getF32Constant(DAG, 0x40c39dad));
- SDValue t10 = DAG.getNode(ISD::FMUL, MVT::f32, t9, X);
- SDValue Log2ofMantissa = DAG.getNode(ISD::FSUB, MVT::f32, t10,
+ SDValue t10 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t9, X);
+ SDValue Log2ofMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t10,
getF32Constant(DAG, 0x4042902c));
- result = DAG.getNode(ISD::FADD, MVT::f32, LogOfExponent, Log2ofMantissa);
+ result = DAG.getNode(ISD::FADD, dl,
+ MVT::f32, LogOfExponent, Log2ofMantissa);
}
} else {
// No special expansion.
- result = DAG.getNode(ISD::FLOG2,
+ result = DAG.getNode(ISD::FLOG2, dl,
getValue(I.getOperand(1)).getValueType(),
getValue(I.getOperand(1)));
}
void
SelectionDAGLowering::visitLog10(CallInst &I) {
SDValue result;
+ DebugLoc dl = getCurDebugLoc();
+
if (getValue(I.getOperand(1)).getValueType() == MVT::f32 &&
LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
SDValue Op = getValue(I.getOperand(1));
- SDValue Op1 = DAG.getNode(ISD::BIT_CONVERT, MVT::i32, Op);
+ SDValue Op1 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, Op);
// Scale the exponent by log10(2) [0.30102999f].
- SDValue Exp = GetExponent(DAG, Op1);
- SDValue LogOfExponent = DAG.getNode(ISD::FMUL, MVT::f32, Exp,
+ SDValue Exp = GetExponent(DAG, Op1, TLI, dl);
+ SDValue LogOfExponent = DAG.getNode(ISD::FMUL, dl, MVT::f32, Exp,
getF32Constant(DAG, 0x3e9a209a));
// Get the significand and build it into a floating-point number with
// exponent of 1.
- SDValue X = GetSignificand(DAG, Op1);
+ SDValue X = GetSignificand(DAG, Op1, dl);
if (LimitFloatPrecision <= 6) {
// For floating-point precision of 6:
- //
+ //
// Log10ofMantissa =
// -0.50419619f +
// (0.60948995f - 0.10380950f * x) * x;
//
// error 0.0014886165, which is 6 bits
- SDValue t0 = DAG.getNode(ISD::FMUL, MVT::f32, X,
+ SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0xbdd49a13));
- SDValue t1 = DAG.getNode(ISD::FADD, MVT::f32, t0,
+ SDValue t1 = DAG.getNode(ISD::FADD, dl, MVT::f32, t0,
getF32Constant(DAG, 0x3f1c0789));
- SDValue t2 = DAG.getNode(ISD::FMUL, MVT::f32, t1, X);
- SDValue Log10ofMantissa = DAG.getNode(ISD::FSUB, MVT::f32, t2,
+ SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
+ SDValue Log10ofMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3f011300));
- result = DAG.getNode(ISD::FADD, MVT::f32, LogOfExponent, Log10ofMantissa);
+ result = DAG.getNode(ISD::FADD, dl,
+ MVT::f32, LogOfExponent, Log10ofMantissa);
} else if (LimitFloatPrecision > 6 && LimitFloatPrecision <= 12) {
// For floating-point precision of 12:
//
// (-0.31664806f + 0.47637168e-1f * x) * x) * x;
//
// error 0.00019228036, which is better than 12 bits
- SDValue t0 = DAG.getNode(ISD::FMUL, MVT::f32, X,
+ SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0x3d431f31));
- SDValue t1 = DAG.getNode(ISD::FSUB, MVT::f32, t0,
+ SDValue t1 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t0,
getF32Constant(DAG, 0x3ea21fb2));
- SDValue t2 = DAG.getNode(ISD::FMUL, MVT::f32, t1, X);
- SDValue t3 = DAG.getNode(ISD::FADD, MVT::f32, t2,
+ SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
+ SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3f6ae232));
- SDValue t4 = DAG.getNode(ISD::FMUL, MVT::f32, t3, X);
- SDValue Log10ofMantissa = DAG.getNode(ISD::FSUB, MVT::f32, t4,
+ SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
+ SDValue Log10ofMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t4,
getF32Constant(DAG, 0x3f25f7c3));
- result = DAG.getNode(ISD::FADD, MVT::f32, LogOfExponent, Log10ofMantissa);
+ result = DAG.getNode(ISD::FADD, dl,
+ MVT::f32, LogOfExponent, Log10ofMantissa);
} else { // LimitFloatPrecision > 12 && LimitFloatPrecision <= 18
// For floating-point precision of 18:
//
// (-0.12539807f + 0.13508273e-1f * x) * x) * x) * x) * x;
//
// error 0.0000037995730, which is better than 18 bits
- SDValue t0 = DAG.getNode(ISD::FMUL, MVT::f32, X,
+ SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0x3c5d51ce));
- SDValue t1 = DAG.getNode(ISD::FSUB, MVT::f32, t0,
+ SDValue t1 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t0,
getF32Constant(DAG, 0x3e00685a));
- SDValue t2 = DAG.getNode(ISD::FMUL, MVT::f32, t1, X);
- SDValue t3 = DAG.getNode(ISD::FADD, MVT::f32, t2,
+ SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
+ SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3efb6798));
- SDValue t4 = DAG.getNode(ISD::FMUL, MVT::f32, t3, X);
- SDValue t5 = DAG.getNode(ISD::FSUB, MVT::f32, t4,
+ SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
+ SDValue t5 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t4,
getF32Constant(DAG, 0x3f88d192));
- SDValue t6 = DAG.getNode(ISD::FMUL, MVT::f32, t5, X);
- SDValue t7 = DAG.getNode(ISD::FADD, MVT::f32, t6,
+ SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
+ SDValue t7 = DAG.getNode(ISD::FADD, dl, MVT::f32, t6,
getF32Constant(DAG, 0x3fc4316c));
- SDValue t8 = DAG.getNode(ISD::FMUL, MVT::f32, t7, X);
- SDValue Log10ofMantissa = DAG.getNode(ISD::FSUB, MVT::f32, t8,
+ SDValue t8 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t7, X);
+ SDValue Log10ofMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t8,
getF32Constant(DAG, 0x3f57ce70));
- result = DAG.getNode(ISD::FADD, MVT::f32, LogOfExponent, Log10ofMantissa);
+ result = DAG.getNode(ISD::FADD, dl,
+ MVT::f32, LogOfExponent, Log10ofMantissa);
}
} else {
// No special expansion.
- result = DAG.getNode(ISD::FLOG10,
+ result = DAG.getNode(ISD::FLOG10, dl,
getValue(I.getOperand(1)).getValueType(),
getValue(I.getOperand(1)));
}
void
SelectionDAGLowering::visitExp2(CallInst &I) {
SDValue result;
+ DebugLoc dl = getCurDebugLoc();
if (getValue(I.getOperand(1)).getValueType() == MVT::f32 &&
LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
SDValue Op = getValue(I.getOperand(1));
- SDValue IntegerPartOfX = DAG.getNode(ISD::FP_TO_SINT, MVT::i32, Op);
+ SDValue IntegerPartOfX = DAG.getNode(ISD::FP_TO_SINT, dl, MVT::i32, Op);
// FractionalPartOfX = x - (float)IntegerPartOfX;
- SDValue t1 = DAG.getNode(ISD::SINT_TO_FP, MVT::f32, IntegerPartOfX);
- SDValue X = DAG.getNode(ISD::FSUB, MVT::f32, Op, t1);
+ SDValue t1 = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::f32, IntegerPartOfX);
+ SDValue X = DAG.getNode(ISD::FSUB, dl, MVT::f32, Op, t1);
// IntegerPartOfX <<= 23;
- IntegerPartOfX = DAG.getNode(ISD::SHL, MVT::i32, IntegerPartOfX,
- DAG.getConstant(23, MVT::i32));
+ IntegerPartOfX = DAG.getNode(ISD::SHL, dl, MVT::i32, IntegerPartOfX,
+ DAG.getConstant(23, TLI.getPointerTy()));
if (LimitFloatPrecision <= 6) {
// For floating-point precision of 6:
- //
+ //
// TwoToFractionalPartOfX =
// 0.997535578f +
// (0.735607626f + 0.252464424f * x) * x;
//
// error 0.0144103317, which is 6 bits
- SDValue t2 = DAG.getNode(ISD::FMUL, MVT::f32, X,
+ SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0x3e814304));
- SDValue t3 = DAG.getNode(ISD::FADD, MVT::f32, t2,
+ SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3f3c50c8));
- SDValue t4 = DAG.getNode(ISD::FMUL, MVT::f32, t3, X);
- SDValue t5 = DAG.getNode(ISD::FADD, MVT::f32, t4,
+ SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
+ SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x3f7f5e7e));
- SDValue t6 = DAG.getNode(ISD::BIT_CONVERT, MVT::i32, t5);
+ SDValue t6 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, t5);
SDValue TwoToFractionalPartOfX =
- DAG.getNode(ISD::ADD, MVT::i32, t6, IntegerPartOfX);
+ DAG.getNode(ISD::ADD, dl, MVT::i32, t6, IntegerPartOfX);
- result = DAG.getNode(ISD::BIT_CONVERT, MVT::f32, TwoToFractionalPartOfX);
+ result = DAG.getNode(ISD::BIT_CONVERT, dl,
+ MVT::f32, TwoToFractionalPartOfX);
} else if (LimitFloatPrecision > 6 && LimitFloatPrecision <= 12) {
// For floating-point precision of 12:
//
// (0.224338339f + 0.792043434e-1f * x) * x) * x;
//
// error 0.000107046256, which is 13 to 14 bits
- SDValue t2 = DAG.getNode(ISD::FMUL, MVT::f32, X,
+ SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0x3da235e3));
- SDValue t3 = DAG.getNode(ISD::FADD, MVT::f32, t2,
+ SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3e65b8f3));
- SDValue t4 = DAG.getNode(ISD::FMUL, MVT::f32, t3, X);
- SDValue t5 = DAG.getNode(ISD::FADD, MVT::f32, t4,
+ SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
+ SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x3f324b07));
- SDValue t6 = DAG.getNode(ISD::FMUL, MVT::f32, t5, X);
- SDValue t7 = DAG.getNode(ISD::FADD, MVT::f32, t6,
+ SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
+ SDValue t7 = DAG.getNode(ISD::FADD, dl, MVT::f32, t6,
getF32Constant(DAG, 0x3f7ff8fd));
- SDValue t8 = DAG.getNode(ISD::BIT_CONVERT, MVT::i32, t7);
+ SDValue t8 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, t7);
SDValue TwoToFractionalPartOfX =
- DAG.getNode(ISD::ADD, MVT::i32, t8, IntegerPartOfX);
+ DAG.getNode(ISD::ADD, dl, MVT::i32, t8, IntegerPartOfX);
- result = DAG.getNode(ISD::BIT_CONVERT, MVT::f32, TwoToFractionalPartOfX);
+ result = DAG.getNode(ISD::BIT_CONVERT, dl,
+ MVT::f32, TwoToFractionalPartOfX);
} else { // LimitFloatPrecision > 12 && LimitFloatPrecision <= 18
// For floating-point precision of 18:
//
// (0.961591928e-2f +
// (0.136028312e-2f + 0.157059148e-3f *x)*x)*x)*x)*x)*x;
// error 2.47208000*10^(-7), which is better than 18 bits
- SDValue t2 = DAG.getNode(ISD::FMUL, MVT::f32, X,
+ SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0x3924b03e));
- SDValue t3 = DAG.getNode(ISD::FADD, MVT::f32, t2,
+ SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3ab24b87));
- SDValue t4 = DAG.getNode(ISD::FMUL, MVT::f32, t3, X);
- SDValue t5 = DAG.getNode(ISD::FADD, MVT::f32, t4,
+ SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
+ SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x3c1d8c17));
- SDValue t6 = DAG.getNode(ISD::FMUL, MVT::f32, t5, X);
- SDValue t7 = DAG.getNode(ISD::FADD, MVT::f32, t6,
+ SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
+ SDValue t7 = DAG.getNode(ISD::FADD, dl, MVT::f32, t6,
getF32Constant(DAG, 0x3d634a1d));
- SDValue t8 = DAG.getNode(ISD::FMUL, MVT::f32, t7, X);
- SDValue t9 = DAG.getNode(ISD::FADD, MVT::f32, t8,
+ SDValue t8 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t7, X);
+ SDValue t9 = DAG.getNode(ISD::FADD, dl, MVT::f32, t8,
getF32Constant(DAG, 0x3e75fe14));
- SDValue t10 = DAG.getNode(ISD::FMUL, MVT::f32, t9, X);
- SDValue t11 = DAG.getNode(ISD::FADD, MVT::f32, t10,
+ SDValue t10 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t9, X);
+ SDValue t11 = DAG.getNode(ISD::FADD, dl, MVT::f32, t10,
getF32Constant(DAG, 0x3f317234));
- SDValue t12 = DAG.getNode(ISD::FMUL, MVT::f32, t11, X);
- SDValue t13 = DAG.getNode(ISD::FADD, MVT::f32, t12,
+ SDValue t12 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t11, X);
+ SDValue t13 = DAG.getNode(ISD::FADD, dl, MVT::f32, t12,
getF32Constant(DAG, 0x3f800000));
- SDValue t14 = DAG.getNode(ISD::BIT_CONVERT, MVT::i32, t13);
+ SDValue t14 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, t13);
SDValue TwoToFractionalPartOfX =
- DAG.getNode(ISD::ADD, MVT::i32, t14, IntegerPartOfX);
+ DAG.getNode(ISD::ADD, dl, MVT::i32, t14, IntegerPartOfX);
- result = DAG.getNode(ISD::BIT_CONVERT, MVT::f32, TwoToFractionalPartOfX);
+ result = DAG.getNode(ISD::BIT_CONVERT, dl,
+ MVT::f32, TwoToFractionalPartOfX);
}
} else {
// No special expansion.
- result = DAG.getNode(ISD::FEXP2,
+ result = DAG.getNode(ISD::FEXP2, dl,
getValue(I.getOperand(1)).getValueType(),
getValue(I.getOperand(1)));
}
SelectionDAGLowering::visitPow(CallInst &I) {
SDValue result;
Value *Val = I.getOperand(1);
+ DebugLoc dl = getCurDebugLoc();
bool IsExp10 = false;
if (getValue(Val).getValueType() == MVT::f32 &&
//
// #define LOG2OF10 3.3219281f
// IntegerPartOfX = (int32_t)(x * LOG2OF10);
- SDValue t0 = DAG.getNode(ISD::FMUL, MVT::f32, Op,
+ SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, Op,
getF32Constant(DAG, 0x40549a78));
- SDValue IntegerPartOfX = DAG.getNode(ISD::FP_TO_SINT, MVT::i32, t0);
+ SDValue IntegerPartOfX = DAG.getNode(ISD::FP_TO_SINT, dl, MVT::i32, t0);
// FractionalPartOfX = x - (float)IntegerPartOfX;
- SDValue t1 = DAG.getNode(ISD::SINT_TO_FP, MVT::f32, IntegerPartOfX);
- SDValue X = DAG.getNode(ISD::FSUB, MVT::f32, t0, t1);
+ SDValue t1 = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::f32, IntegerPartOfX);
+ SDValue X = DAG.getNode(ISD::FSUB, dl, MVT::f32, t0, t1);
// IntegerPartOfX <<= 23;
- IntegerPartOfX = DAG.getNode(ISD::SHL, MVT::i32, IntegerPartOfX,
- DAG.getConstant(23, MVT::i32));
+ IntegerPartOfX = DAG.getNode(ISD::SHL, dl, MVT::i32, IntegerPartOfX,
+ DAG.getConstant(23, TLI.getPointerTy()));
if (LimitFloatPrecision <= 6) {
// For floating-point precision of 6:
- //
+ //
// twoToFractionalPartOfX =
// 0.997535578f +
// (0.735607626f + 0.252464424f * x) * x;
- //
+ //
// error 0.0144103317, which is 6 bits
- SDValue t2 = DAG.getNode(ISD::FMUL, MVT::f32, X,
+ SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0x3e814304));
- SDValue t3 = DAG.getNode(ISD::FADD, MVT::f32, t2,
+ SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3f3c50c8));
- SDValue t4 = DAG.getNode(ISD::FMUL, MVT::f32, t3, X);
- SDValue t5 = DAG.getNode(ISD::FADD, MVT::f32, t4,
+ SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
+ SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x3f7f5e7e));
- SDValue t6 = DAG.getNode(ISD::BIT_CONVERT, MVT::i32, t5);
+ SDValue t6 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, t5);
SDValue TwoToFractionalPartOfX =
- DAG.getNode(ISD::ADD, MVT::i32, t6, IntegerPartOfX);
+ DAG.getNode(ISD::ADD, dl, MVT::i32, t6, IntegerPartOfX);
- result = DAG.getNode(ISD::BIT_CONVERT, MVT::f32, TwoToFractionalPartOfX);
+ result = DAG.getNode(ISD::BIT_CONVERT, dl,
+ MVT::f32, TwoToFractionalPartOfX);
} else if (LimitFloatPrecision > 6 && LimitFloatPrecision <= 12) {
// For floating-point precision of 12:
//
// (0.224338339f + 0.792043434e-1f * x) * x) * x;
//
// error 0.000107046256, which is 13 to 14 bits
- SDValue t2 = DAG.getNode(ISD::FMUL, MVT::f32, X,
+ SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0x3da235e3));
- SDValue t3 = DAG.getNode(ISD::FADD, MVT::f32, t2,
+ SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3e65b8f3));
- SDValue t4 = DAG.getNode(ISD::FMUL, MVT::f32, t3, X);
- SDValue t5 = DAG.getNode(ISD::FADD, MVT::f32, t4,
+ SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
+ SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x3f324b07));
- SDValue t6 = DAG.getNode(ISD::FMUL, MVT::f32, t5, X);
- SDValue t7 = DAG.getNode(ISD::FADD, MVT::f32, t6,
+ SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
+ SDValue t7 = DAG.getNode(ISD::FADD, dl, MVT::f32, t6,
getF32Constant(DAG, 0x3f7ff8fd));
- SDValue t8 = DAG.getNode(ISD::BIT_CONVERT, MVT::i32, t7);
+ SDValue t8 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, t7);
SDValue TwoToFractionalPartOfX =
- DAG.getNode(ISD::ADD, MVT::i32, t8, IntegerPartOfX);
+ DAG.getNode(ISD::ADD, dl, MVT::i32, t8, IntegerPartOfX);
- result = DAG.getNode(ISD::BIT_CONVERT, MVT::f32, TwoToFractionalPartOfX);
+ result = DAG.getNode(ISD::BIT_CONVERT, dl,
+ MVT::f32, TwoToFractionalPartOfX);
} else { // LimitFloatPrecision > 12 && LimitFloatPrecision <= 18
// For floating-point precision of 18:
//
// (0.961591928e-2f +
// (0.136028312e-2f + 0.157059148e-3f *x)*x)*x)*x)*x)*x;
// error 2.47208000*10^(-7), which is better than 18 bits
- SDValue t2 = DAG.getNode(ISD::FMUL, MVT::f32, X,
+ SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
getF32Constant(DAG, 0x3924b03e));
- SDValue t3 = DAG.getNode(ISD::FADD, MVT::f32, t2,
+ SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2,
getF32Constant(DAG, 0x3ab24b87));
- SDValue t4 = DAG.getNode(ISD::FMUL, MVT::f32, t3, X);
- SDValue t5 = DAG.getNode(ISD::FADD, MVT::f32, t4,
+ SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
+ SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
getF32Constant(DAG, 0x3c1d8c17));
- SDValue t6 = DAG.getNode(ISD::FMUL, MVT::f32, t5, X);
- SDValue t7 = DAG.getNode(ISD::FADD, MVT::f32, t6,
+ SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
+ SDValue t7 = DAG.getNode(ISD::FADD, dl, MVT::f32, t6,
getF32Constant(DAG, 0x3d634a1d));
- SDValue t8 = DAG.getNode(ISD::FMUL, MVT::f32, t7, X);
- SDValue t9 = DAG.getNode(ISD::FADD, MVT::f32, t8,
+ SDValue t8 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t7, X);
+ SDValue t9 = DAG.getNode(ISD::FADD, dl, MVT::f32, t8,
getF32Constant(DAG, 0x3e75fe14));
- SDValue t10 = DAG.getNode(ISD::FMUL, MVT::f32, t9, X);
- SDValue t11 = DAG.getNode(ISD::FADD, MVT::f32, t10,
+ SDValue t10 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t9, X);
+ SDValue t11 = DAG.getNode(ISD::FADD, dl, MVT::f32, t10,
getF32Constant(DAG, 0x3f317234));
- SDValue t12 = DAG.getNode(ISD::FMUL, MVT::f32, t11, X);
- SDValue t13 = DAG.getNode(ISD::FADD, MVT::f32, t12,
+ SDValue t12 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t11, X);
+ SDValue t13 = DAG.getNode(ISD::FADD, dl, MVT::f32, t12,
getF32Constant(DAG, 0x3f800000));
- SDValue t14 = DAG.getNode(ISD::BIT_CONVERT, MVT::i32, t13);
+ SDValue t14 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, t13);
SDValue TwoToFractionalPartOfX =
- DAG.getNode(ISD::ADD, MVT::i32, t14, IntegerPartOfX);
+ DAG.getNode(ISD::ADD, dl, MVT::i32, t14, IntegerPartOfX);
- result = DAG.getNode(ISD::BIT_CONVERT, MVT::f32, TwoToFractionalPartOfX);
+ result = DAG.getNode(ISD::BIT_CONVERT, dl,
+ MVT::f32, TwoToFractionalPartOfX);
}
} else {
// No special expansion.
- result = DAG.getNode(ISD::FPOW,
+ result = DAG.getNode(ISD::FPOW, dl,
getValue(I.getOperand(1)).getValueType(),
getValue(I.getOperand(1)),
getValue(I.getOperand(2)));
/// otherwise lower it and return null.
const char *
SelectionDAGLowering::visitIntrinsicCall(CallInst &I, unsigned Intrinsic) {
+ DebugLoc dl = getCurDebugLoc();
switch (Intrinsic) {
default:
// By default, turn this into a target intrinsic node.
case Intrinsic::vaend: visitVAEnd(I); return 0;
case Intrinsic::vacopy: visitVACopy(I); return 0;
case Intrinsic::returnaddress:
- setValue(&I, DAG.getNode(ISD::RETURNADDR, TLI.getPointerTy(),
+ setValue(&I, DAG.getNode(ISD::RETURNADDR, dl, TLI.getPointerTy(),
getValue(I.getOperand(1))));
return 0;
case Intrinsic::frameaddress:
- setValue(&I, DAG.getNode(ISD::FRAMEADDR, TLI.getPointerTy(),
+ setValue(&I, DAG.getNode(ISD::FRAMEADDR, dl, TLI.getPointerTy(),
getValue(I.getOperand(1))));
return 0;
case Intrinsic::setjmp:
case Intrinsic::longjmp:
return "_longjmp"+!TLI.usesUnderscoreLongJmp();
break;
- case Intrinsic::memcpy_i32:
- case Intrinsic::memcpy_i64: {
+ case Intrinsic::memcpy: {
SDValue Op1 = getValue(I.getOperand(1));
SDValue Op2 = getValue(I.getOperand(2));
SDValue Op3 = getValue(I.getOperand(3));
unsigned Align = cast<ConstantInt>(I.getOperand(4))->getZExtValue();
- DAG.setRoot(DAG.getMemcpy(getRoot(), Op1, Op2, Op3, Align, false,
+ DAG.setRoot(DAG.getMemcpy(getRoot(), dl, Op1, Op2, Op3, Align, false,
I.getOperand(1), 0, I.getOperand(2), 0));
return 0;
}
- case Intrinsic::memset_i32:
- case Intrinsic::memset_i64: {
+ case Intrinsic::memset: {
SDValue Op1 = getValue(I.getOperand(1));
SDValue Op2 = getValue(I.getOperand(2));
SDValue Op3 = getValue(I.getOperand(3));
unsigned Align = cast<ConstantInt>(I.getOperand(4))->getZExtValue();
- DAG.setRoot(DAG.getMemset(getRoot(), Op1, Op2, Op3, Align,
+ DAG.setRoot(DAG.getMemset(getRoot(), dl, Op1, Op2, Op3, Align,
I.getOperand(1), 0));
return 0;
}
- case Intrinsic::memmove_i32:
- case Intrinsic::memmove_i64: {
+ case Intrinsic::memmove: {
SDValue Op1 = getValue(I.getOperand(1));
SDValue Op2 = getValue(I.getOperand(2));
SDValue Op3 = getValue(I.getOperand(3));
Size = C->getZExtValue();
if (AA->alias(I.getOperand(1), Size, I.getOperand(2), Size) ==
AliasAnalysis::NoAlias) {
- DAG.setRoot(DAG.getMemcpy(getRoot(), Op1, Op2, Op3, Align, false,
+ DAG.setRoot(DAG.getMemcpy(getRoot(), dl, Op1, Op2, Op3, Align, false,
I.getOperand(1), 0, I.getOperand(2), 0));
return 0;
}
- DAG.setRoot(DAG.getMemmove(getRoot(), Op1, Op2, Op3, Align,
+ DAG.setRoot(DAG.getMemmove(getRoot(), dl, Op1, Op2, Op3, Align,
I.getOperand(1), 0, I.getOperand(2), 0));
return 0;
}
case Intrinsic::dbg_stoppoint: {
- MachineModuleInfo *MMI = DAG.getMachineModuleInfo();
+ DwarfWriter *DW = DAG.getDwarfWriter();
DbgStopPointInst &SPI = cast<DbgStopPointInst>(I);
- if (MMI && SPI.getContext() && MMI->Verify(SPI.getContext())) {
- DebugInfoDesc *DD = MMI->getDescFor(SPI.getContext());
- assert(DD && "Not a debug information descriptor");
+ if (DW && DW->ValidDebugInfo(SPI.getContext())) {
DAG.setRoot(DAG.getDbgStopPoint(getRoot(),
SPI.getLine(),
SPI.getColumn(),
- cast<CompileUnitDesc>(DD)));
+ SPI.getContext()));
+ DICompileUnit CU(cast<GlobalVariable>(SPI.getContext()));
+ unsigned SrcFile = DW->RecordSource(CU.getDirectory(), CU.getFilename());
+ unsigned idx = DAG.getMachineFunction().
+ getOrCreateDebugLocID(SrcFile,
+ SPI.getLine(),
+ SPI.getColumn());
+ setCurDebugLoc(DebugLoc::get(idx));
}
-
return 0;
}
case Intrinsic::dbg_region_start: {
- MachineModuleInfo *MMI = DAG.getMachineModuleInfo();
+ DwarfWriter *DW = DAG.getDwarfWriter();
DbgRegionStartInst &RSI = cast<DbgRegionStartInst>(I);
- if (MMI && RSI.getContext() && MMI->Verify(RSI.getContext())) {
- unsigned LabelID = MMI->RecordRegionStart(RSI.getContext());
- DAG.setRoot(DAG.getLabel(ISD::DBG_LABEL, getRoot(), LabelID));
+ if (DW && DW->ValidDebugInfo(RSI.getContext())) {
+ unsigned LabelID =
+ DW->RecordRegionStart(cast<GlobalVariable>(RSI.getContext()));
+ DAG.setRoot(DAG.getLabel(ISD::DBG_LABEL, getCurDebugLoc(),
+ getRoot(), LabelID));
}
return 0;
}
case Intrinsic::dbg_region_end: {
- MachineModuleInfo *MMI = DAG.getMachineModuleInfo();
+ DwarfWriter *DW = DAG.getDwarfWriter();
DbgRegionEndInst &REI = cast<DbgRegionEndInst>(I);
- if (MMI && REI.getContext() && MMI->Verify(REI.getContext())) {
- unsigned LabelID = MMI->RecordRegionEnd(REI.getContext());
- DAG.setRoot(DAG.getLabel(ISD::DBG_LABEL, getRoot(), LabelID));
+ if (DW && DW->ValidDebugInfo(REI.getContext())) {
+ unsigned LabelID =
+ DW->RecordRegionEnd(cast<GlobalVariable>(REI.getContext()));
+ DAG.setRoot(DAG.getLabel(ISD::DBG_LABEL, getCurDebugLoc(),
+ getRoot(), LabelID));
}
return 0;
}
case Intrinsic::dbg_func_start: {
- MachineModuleInfo *MMI = DAG.getMachineModuleInfo();
- if (!MMI) return 0;
+ DwarfWriter *DW = DAG.getDwarfWriter();
+ if (!DW) return 0;
DbgFuncStartInst &FSI = cast<DbgFuncStartInst>(I);
Value *SP = FSI.getSubprogram();
- if (SP && MMI->Verify(SP)) {
+ if (SP && DW->ValidDebugInfo(SP)) {
// llvm.dbg.func.start implicitly defines a dbg_stoppoint which is
// what (most?) gdb expects.
- DebugInfoDesc *DD = MMI->getDescFor(SP);
- assert(DD && "Not a debug information descriptor");
- SubprogramDesc *Subprogram = cast<SubprogramDesc>(DD);
- const CompileUnitDesc *CompileUnit = Subprogram->getFile();
- unsigned SrcFile = MMI->RecordSource(CompileUnit);
- // Record the source line but does create a label. It will be emitted
- // at asm emission time.
- MMI->RecordSourceLine(Subprogram->getLine(), 0, SrcFile);
+ DISubprogram Subprogram(cast<GlobalVariable>(SP));
+ DICompileUnit CompileUnit = Subprogram.getCompileUnit();
+ unsigned SrcFile = DW->RecordSource(CompileUnit.getDirectory(),
+ CompileUnit.getFilename());
+
+ // Record the source line but does not create a label for the normal
+ // function start. It will be emitted at asm emission time. However,
+ // create a label if this is a beginning of inlined function.
+ unsigned Line = Subprogram.getLineNumber();
+ unsigned LabelID = DW->RecordSourceLine(Line, 0, SrcFile);
+
+ if (DW->getRecordSourceLineCount() != 1)
+ DAG.setRoot(DAG.getLabel(ISD::DBG_LABEL, getCurDebugLoc(),
+ getRoot(), LabelID));
+
+ setCurDebugLoc(DebugLoc::get(DAG.getMachineFunction().
+ getOrCreateDebugLocID(SrcFile, Line, 0)));
}
return 0;
}
case Intrinsic::dbg_declare: {
- MachineModuleInfo *MMI = DAG.getMachineModuleInfo();
+ DwarfWriter *DW = DAG.getDwarfWriter();
DbgDeclareInst &DI = cast<DbgDeclareInst>(I);
Value *Variable = DI.getVariable();
- if (MMI && Variable && MMI->Verify(Variable))
- DAG.setRoot(DAG.getNode(ISD::DECLARE, MVT::Other, getRoot(),
+ if (DW && DW->ValidDebugInfo(Variable))
+ DAG.setRoot(DAG.getNode(ISD::DECLARE, dl, MVT::Other, getRoot(),
getValue(DI.getAddress()), getValue(Variable)));
return 0;
}
-
+
case Intrinsic::eh_exception: {
if (!CurMBB->isLandingPad()) {
// FIXME: Mark exception register as live in. Hack for PR1508.
SDVTList VTs = DAG.getVTList(TLI.getPointerTy(), MVT::Other);
SDValue Ops[1];
Ops[0] = DAG.getRoot();
- SDValue Op = DAG.getNode(ISD::EXCEPTIONADDR, VTs, Ops, 1);
+ SDValue Op = DAG.getNode(ISD::EXCEPTIONADDR, dl, VTs, Ops, 1);
setValue(&I, Op);
DAG.setRoot(Op.getValue(1));
return 0;
MachineModuleInfo *MMI = DAG.getMachineModuleInfo();
MVT VT = (Intrinsic == Intrinsic::eh_selector_i32 ?
MVT::i32 : MVT::i64);
-
+
if (MMI) {
if (CurMBB->isLandingPad())
AddCatchInfo(I, MMI, CurMBB);
SDValue Ops[2];
Ops[0] = getValue(I.getOperand(1));
Ops[1] = getRoot();
- SDValue Op = DAG.getNode(ISD::EHSELECTION, VTs, Ops, 2);
+ SDValue Op = DAG.getNode(ISD::EHSELECTION, dl, VTs, Ops, 2);
setValue(&I, Op);
DAG.setRoot(Op.getValue(1));
} else {
setValue(&I, DAG.getConstant(0, VT));
}
-
+
return 0;
}
case Intrinsic::eh_return_i64:
if (MachineModuleInfo *MMI = DAG.getMachineModuleInfo()) {
MMI->setCallsEHReturn(true);
- DAG.setRoot(DAG.getNode(ISD::EH_RETURN,
+ DAG.setRoot(DAG.getNode(ISD::EH_RETURN, dl,
MVT::Other,
getControlRoot(),
getValue(I.getOperand(1)),
MVT VT = getValue(I.getOperand(1)).getValueType();
SDValue CfaArg;
if (VT.bitsGT(TLI.getPointerTy()))
- CfaArg = DAG.getNode(ISD::TRUNCATE,
+ CfaArg = DAG.getNode(ISD::TRUNCATE, dl,
TLI.getPointerTy(), getValue(I.getOperand(1)));
else
- CfaArg = DAG.getNode(ISD::SIGN_EXTEND,
+ CfaArg = DAG.getNode(ISD::SIGN_EXTEND, dl,
TLI.getPointerTy(), getValue(I.getOperand(1)));
- SDValue Offset = DAG.getNode(ISD::ADD,
+ SDValue Offset = DAG.getNode(ISD::ADD, dl,
TLI.getPointerTy(),
- DAG.getNode(ISD::FRAME_TO_ARGS_OFFSET,
+ DAG.getNode(ISD::FRAME_TO_ARGS_OFFSET, dl,
TLI.getPointerTy()),
CfaArg);
- setValue(&I, DAG.getNode(ISD::ADD,
+ setValue(&I, DAG.getNode(ISD::ADD, dl,
TLI.getPointerTy(),
- DAG.getNode(ISD::FRAMEADDR,
+ DAG.getNode(ISD::FRAMEADDR, dl,
TLI.getPointerTy(),
DAG.getConstant(0,
TLI.getPointerTy())),
return 0;
}
+ case Intrinsic::convertff:
+ case Intrinsic::convertfsi:
+ case Intrinsic::convertfui:
+ case Intrinsic::convertsif:
+ case Intrinsic::convertuif:
+ case Intrinsic::convertss:
+ case Intrinsic::convertsu:
+ case Intrinsic::convertus:
+ case Intrinsic::convertuu: {
+ ISD::CvtCode Code = ISD::CVT_INVALID;
+ switch (Intrinsic) {
+ 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::convertsif: Code = ISD::CVT_SF; break;
+ case Intrinsic::convertuif: Code = ISD::CVT_UF; break;
+ case Intrinsic::convertss: Code = ISD::CVT_SS; break;
+ case Intrinsic::convertsu: Code = ISD::CVT_SU; break;
+ case Intrinsic::convertus: Code = ISD::CVT_US; break;
+ case Intrinsic::convertuu: Code = ISD::CVT_UU; break;
+ }
+ MVT DestVT = TLI.getValueType(I.getType());
+ Value* Op1 = I.getOperand(1);
+ setValue(&I, DAG.getConvertRndSat(DestVT, getCurDebugLoc(), getValue(Op1),
+ DAG.getValueType(DestVT),
+ DAG.getValueType(getValue(Op1).getValueType()),
+ getValue(I.getOperand(2)),
+ getValue(I.getOperand(3)),
+ Code));
+ return 0;
+ }
+
case Intrinsic::sqrt:
- setValue(&I, DAG.getNode(ISD::FSQRT,
+ setValue(&I, DAG.getNode(ISD::FSQRT, dl,
getValue(I.getOperand(1)).getValueType(),
getValue(I.getOperand(1))));
return 0;
case Intrinsic::powi:
- setValue(&I, DAG.getNode(ISD::FPOWI,
+ setValue(&I, DAG.getNode(ISD::FPOWI, dl,
getValue(I.getOperand(1)).getValueType(),
getValue(I.getOperand(1)),
getValue(I.getOperand(2))));
return 0;
case Intrinsic::sin:
- setValue(&I, DAG.getNode(ISD::FSIN,
+ setValue(&I, DAG.getNode(ISD::FSIN, dl,
getValue(I.getOperand(1)).getValueType(),
getValue(I.getOperand(1))));
return 0;
case Intrinsic::cos:
- setValue(&I, DAG.getNode(ISD::FCOS,
+ setValue(&I, DAG.getNode(ISD::FCOS, dl,
getValue(I.getOperand(1)).getValueType(),
getValue(I.getOperand(1))));
return 0;
return 0;
case Intrinsic::pcmarker: {
SDValue Tmp = getValue(I.getOperand(1));
- DAG.setRoot(DAG.getNode(ISD::PCMARKER, MVT::Other, getRoot(), Tmp));
+ DAG.setRoot(DAG.getNode(ISD::PCMARKER, dl, MVT::Other, getRoot(), Tmp));
return 0;
}
case Intrinsic::readcyclecounter: {
SDValue Op = getRoot();
- SDValue Tmp = DAG.getNode(ISD::READCYCLECOUNTER,
+ SDValue Tmp = DAG.getNode(ISD::READCYCLECOUNTER, dl,
DAG.getNodeValueTypes(MVT::i64, MVT::Other), 2,
&Op, 1);
setValue(&I, Tmp);
abort();
}
case Intrinsic::bswap:
- setValue(&I, DAG.getNode(ISD::BSWAP,
+ setValue(&I, DAG.getNode(ISD::BSWAP, dl,
getValue(I.getOperand(1)).getValueType(),
getValue(I.getOperand(1))));
return 0;
case Intrinsic::cttz: {
SDValue Arg = getValue(I.getOperand(1));
MVT Ty = Arg.getValueType();
- SDValue result = DAG.getNode(ISD::CTTZ, Ty, Arg);
+ SDValue result = DAG.getNode(ISD::CTTZ, dl, Ty, Arg);
setValue(&I, result);
return 0;
}
case Intrinsic::ctlz: {
SDValue Arg = getValue(I.getOperand(1));
MVT Ty = Arg.getValueType();
- SDValue result = DAG.getNode(ISD::CTLZ, Ty, Arg);
+ SDValue result = DAG.getNode(ISD::CTLZ, dl, Ty, Arg);
setValue(&I, result);
return 0;
}
case Intrinsic::ctpop: {
SDValue Arg = getValue(I.getOperand(1));
MVT Ty = Arg.getValueType();
- SDValue result = DAG.getNode(ISD::CTPOP, Ty, Arg);
+ SDValue result = DAG.getNode(ISD::CTPOP, dl, Ty, Arg);
setValue(&I, result);
return 0;
}
case Intrinsic::stacksave: {
SDValue Op = getRoot();
- SDValue Tmp = DAG.getNode(ISD::STACKSAVE,
+ SDValue Tmp = DAG.getNode(ISD::STACKSAVE, dl,
DAG.getNodeValueTypes(TLI.getPointerTy(), MVT::Other), 2, &Op, 1);
setValue(&I, Tmp);
DAG.setRoot(Tmp.getValue(1));
}
case Intrinsic::stackrestore: {
SDValue Tmp = getValue(I.getOperand(1));
- DAG.setRoot(DAG.getNode(ISD::STACKRESTORE, MVT::Other, getRoot(), Tmp));
+ DAG.setRoot(DAG.getNode(ISD::STACKRESTORE, dl, MVT::Other, getRoot(), Tmp));
+ return 0;
+ }
+ case Intrinsic::stackprotector: {
+ // Emit code into the DAG to store the stack guard onto the stack.
+ MachineFunction &MF = DAG.getMachineFunction();
+ MachineFrameInfo *MFI = MF.getFrameInfo();
+ MVT PtrTy = TLI.getPointerTy();
+
+ SDValue Src = getValue(I.getOperand(1)); // The guard's value.
+ AllocaInst *Slot = cast<AllocaInst>(I.getOperand(2));
+
+ int FI = FuncInfo.StaticAllocaMap[Slot];
+ MFI->setStackProtectorIndex(FI);
+
+ SDValue FIN = DAG.getFrameIndex(FI, PtrTy);
+
+ // Store the stack protector onto the stack.
+ SDValue Result = DAG.getStore(getRoot(), getCurDebugLoc(), Src, FIN,
+ PseudoSourceValue::getFixedStack(FI),
+ 0, true);
+ setValue(&I, Result);
+ DAG.setRoot(Result);
return 0;
}
case Intrinsic::var_annotation:
Ops[4] = DAG.getSrcValue(I.getOperand(1));
Ops[5] = DAG.getSrcValue(F);
- SDValue Tmp = DAG.getNode(ISD::TRAMPOLINE,
+ SDValue Tmp = DAG.getNode(ISD::TRAMPOLINE, dl,
DAG.getNodeValueTypes(TLI.getPointerTy(),
MVT::Other), 2,
Ops, 6);
if (GFI) {
Value *Alloca = I.getOperand(1);
Constant *TypeMap = cast<Constant>(I.getOperand(2));
-
+
FrameIndexSDNode *FI = cast<FrameIndexSDNode>(getValue(Alloca).getNode());
GFI->addStackRoot(FI->getIndex(), TypeMap);
}
return 0;
case Intrinsic::flt_rounds: {
- setValue(&I, DAG.getNode(ISD::FLT_ROUNDS_, MVT::i32));
+ setValue(&I, DAG.getNode(ISD::FLT_ROUNDS_, dl, MVT::i32));
return 0;
}
case Intrinsic::trap: {
- DAG.setRoot(DAG.getNode(ISD::TRAP, MVT::Other, getRoot()));
+ DAG.setRoot(DAG.getNode(ISD::TRAP, 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::prefetch: {
SDValue Ops[4];
Ops[0] = getRoot();
Ops[1] = getValue(I.getOperand(1));
Ops[2] = getValue(I.getOperand(2));
Ops[3] = getValue(I.getOperand(3));
- DAG.setRoot(DAG.getNode(ISD::PREFETCH, MVT::Other, &Ops[0], 4));
+ DAG.setRoot(DAG.getNode(ISD::PREFETCH, dl, MVT::Other, &Ops[0], 4));
return 0;
}
-
+
case Intrinsic::memory_barrier: {
SDValue Ops[6];
Ops[0] = getRoot();
for (int x = 1; x < 6; ++x)
Ops[x] = getValue(I.getOperand(x));
- DAG.setRoot(DAG.getNode(ISD::MEMBARRIER, MVT::Other, &Ops[0], 6));
+ DAG.setRoot(DAG.getNode(ISD::MEMBARRIER, dl, MVT::Other, &Ops[0], 6));
return 0;
}
case Intrinsic::atomic_cmp_swap: {
- SDValue Root = getRoot();
- SDValue L;
- switch (getValue(I.getOperand(2)).getValueType().getSimpleVT()) {
- case MVT::i8:
- L = DAG.getAtomic(ISD::ATOMIC_CMP_SWAP_8, Root,
- getValue(I.getOperand(1)),
- getValue(I.getOperand(2)),
- getValue(I.getOperand(3)),
- I.getOperand(1));
- break;
- case MVT::i16:
- L = DAG.getAtomic(ISD::ATOMIC_CMP_SWAP_16, Root,
- getValue(I.getOperand(1)),
- getValue(I.getOperand(2)),
- getValue(I.getOperand(3)),
- I.getOperand(1));
- break;
- case MVT::i32:
- L = DAG.getAtomic(ISD::ATOMIC_CMP_SWAP_32, Root,
- getValue(I.getOperand(1)),
- getValue(I.getOperand(2)),
- getValue(I.getOperand(3)),
- I.getOperand(1));
- break;
- case MVT::i64:
- L = DAG.getAtomic(ISD::ATOMIC_CMP_SWAP_64, Root,
- getValue(I.getOperand(1)),
- getValue(I.getOperand(2)),
- getValue(I.getOperand(3)),
- I.getOperand(1));
- break;
- default:
- assert(0 && "Invalid atomic type");
- abort();
- }
+ SDValue Root = getRoot();
+ SDValue L =
+ DAG.getAtomic(ISD::ATOMIC_CMP_SWAP, getCurDebugLoc(),
+ getValue(I.getOperand(2)).getValueType().getSimpleVT(),
+ Root,
+ getValue(I.getOperand(1)),
+ getValue(I.getOperand(2)),
+ getValue(I.getOperand(3)),
+ I.getOperand(1));
setValue(&I, L);
DAG.setRoot(L.getValue(1));
return 0;
}
case Intrinsic::atomic_load_add:
- switch (getValue(I.getOperand(2)).getValueType().getSimpleVT()) {
- case MVT::i8:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_ADD_8);
- case MVT::i16:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_ADD_16);
- case MVT::i32:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_ADD_32);
- case MVT::i64:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_ADD_64);
- default:
- assert(0 && "Invalid atomic type");
- abort();
- }
+ return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_ADD);
case Intrinsic::atomic_load_sub:
- switch (getValue(I.getOperand(2)).getValueType().getSimpleVT()) {
- case MVT::i8:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_SUB_8);
- case MVT::i16:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_SUB_16);
- case MVT::i32:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_SUB_32);
- case MVT::i64:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_SUB_64);
- default:
- assert(0 && "Invalid atomic type");
- abort();
- }
+ return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_SUB);
case Intrinsic::atomic_load_or:
- switch (getValue(I.getOperand(2)).getValueType().getSimpleVT()) {
- case MVT::i8:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_OR_8);
- case MVT::i16:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_OR_16);
- case MVT::i32:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_OR_32);
- case MVT::i64:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_OR_64);
- default:
- assert(0 && "Invalid atomic type");
- abort();
- }
+ return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_OR);
case Intrinsic::atomic_load_xor:
- switch (getValue(I.getOperand(2)).getValueType().getSimpleVT()) {
- case MVT::i8:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_XOR_8);
- case MVT::i16:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_XOR_16);
- case MVT::i32:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_XOR_32);
- case MVT::i64:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_XOR_64);
- default:
- assert(0 && "Invalid atomic type");
- abort();
- }
+ return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_XOR);
case Intrinsic::atomic_load_and:
- switch (getValue(I.getOperand(2)).getValueType().getSimpleVT()) {
- case MVT::i8:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_AND_8);
- case MVT::i16:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_AND_16);
- case MVT::i32:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_AND_32);
- case MVT::i64:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_AND_64);
- default:
- assert(0 && "Invalid atomic type");
- abort();
- }
+ return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_AND);
case Intrinsic::atomic_load_nand:
- switch (getValue(I.getOperand(2)).getValueType().getSimpleVT()) {
- case MVT::i8:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_NAND_8);
- case MVT::i16:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_NAND_16);
- case MVT::i32:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_NAND_32);
- case MVT::i64:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_NAND_64);
- default:
- assert(0 && "Invalid atomic type");
- abort();
- }
+ return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_NAND);
case Intrinsic::atomic_load_max:
- switch (getValue(I.getOperand(2)).getValueType().getSimpleVT()) {
- case MVT::i8:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_MAX_8);
- case MVT::i16:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_MAX_16);
- case MVT::i32:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_MAX_32);
- case MVT::i64:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_MAX_64);
- default:
- assert(0 && "Invalid atomic type");
- abort();
- }
+ return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_MAX);
case Intrinsic::atomic_load_min:
- switch (getValue(I.getOperand(2)).getValueType().getSimpleVT()) {
- case MVT::i8:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_MIN_8);
- case MVT::i16:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_MIN_16);
- case MVT::i32:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_MIN_32);
- case MVT::i64:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_MIN_64);
- default:
- assert(0 && "Invalid atomic type");
- abort();
- }
+ return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_MIN);
case Intrinsic::atomic_load_umin:
- switch (getValue(I.getOperand(2)).getValueType().getSimpleVT()) {
- case MVT::i8:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_UMIN_8);
- case MVT::i16:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_UMIN_16);
- case MVT::i32:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_UMIN_32);
- case MVT::i64:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_UMIN_64);
- default:
- assert(0 && "Invalid atomic type");
- abort();
- }
+ return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_UMIN);
case Intrinsic::atomic_load_umax:
- switch (getValue(I.getOperand(2)).getValueType().getSimpleVT()) {
- case MVT::i8:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_UMAX_8);
- case MVT::i16:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_UMAX_16);
- case MVT::i32:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_UMAX_32);
- case MVT::i64:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_UMAX_64);
- default:
- assert(0 && "Invalid atomic type");
- abort();
- }
+ return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_UMAX);
case Intrinsic::atomic_swap:
- switch (getValue(I.getOperand(2)).getValueType().getSimpleVT()) {
- case MVT::i8:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_SWAP_8);
- case MVT::i16:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_SWAP_16);
- case MVT::i32:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_SWAP_32);
- case MVT::i64:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_SWAP_64);
- default:
- assert(0 && "Invalid atomic type");
- abort();
- }
+ return implVisitBinaryAtomic(I, ISD::ATOMIC_SWAP);
}
}
Entry.Node = ArgNode; Entry.Ty = (*i)->getType();
unsigned attrInd = i - CS.arg_begin() + 1;
- Entry.isSExt = CS.paramHasAttr(attrInd, ParamAttr::SExt);
- Entry.isZExt = CS.paramHasAttr(attrInd, ParamAttr::ZExt);
- Entry.isInReg = CS.paramHasAttr(attrInd, ParamAttr::InReg);
- Entry.isSRet = CS.paramHasAttr(attrInd, ParamAttr::StructRet);
- Entry.isNest = CS.paramHasAttr(attrInd, ParamAttr::Nest);
- Entry.isByVal = CS.paramHasAttr(attrInd, ParamAttr::ByVal);
+ Entry.isSExt = CS.paramHasAttr(attrInd, Attribute::SExt);
+ Entry.isZExt = CS.paramHasAttr(attrInd, Attribute::ZExt);
+ Entry.isInReg = CS.paramHasAttr(attrInd, Attribute::InReg);
+ Entry.isSRet = CS.paramHasAttr(attrInd, Attribute::StructRet);
+ Entry.isNest = CS.paramHasAttr(attrInd, Attribute::Nest);
+ Entry.isByVal = CS.paramHasAttr(attrInd, Attribute::ByVal);
Entry.Alignment = CS.getParamAlignment(attrInd);
Args.push_back(Entry);
}
// Both PendingLoads and PendingExports must be flushed here;
// this call might not return.
(void)getRoot();
- DAG.setRoot(DAG.getLabel(ISD::EH_LABEL, getControlRoot(), BeginLabel));
+ DAG.setRoot(DAG.getLabel(ISD::EH_LABEL, getCurDebugLoc(),
+ getControlRoot(), BeginLabel));
}
std::pair<SDValue,SDValue> Result =
TLI.LowerCallTo(getRoot(), CS.getType(),
- CS.paramHasAttr(0, ParamAttr::SExt),
- CS.paramHasAttr(0, ParamAttr::ZExt),
- FTy->isVarArg(), CS.getCallingConv(),
+ CS.paramHasAttr(0, Attribute::SExt),
+ CS.paramHasAttr(0, Attribute::ZExt), FTy->isVarArg(),
+ CS.paramHasAttr(0, Attribute::InReg),
+ CS.getCallingConv(),
IsTailCall && PerformTailCallOpt,
- Callee, Args, DAG);
+ Callee, Args, DAG, getCurDebugLoc());
if (CS.getType() != Type::VoidTy)
setValue(CS.getInstruction(), Result.first);
DAG.setRoot(Result.second);
// 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.
EndLabel = MMI->NextLabelID();
- DAG.setRoot(DAG.getLabel(ISD::EH_LABEL, getRoot(), EndLabel));
+ DAG.setRoot(DAG.getLabel(ISD::EH_LABEL, getCurDebugLoc(),
+ getRoot(), EndLabel));
// Inform MachineModuleInfo of range.
MMI->addInvoke(LandingPad, BeginLabel, EndLabel);
const char *RenameFn = 0;
if (Function *F = I.getCalledFunction()) {
if (F->isDeclaration()) {
+ const TargetIntrinsicInfo *II = TLI.getTargetMachine().getIntrinsicInfo();
+ if (II) {
+ if (unsigned IID = II->getIntrinsicID(F)) {
+ RenameFn = visitIntrinsicCall(I, IID);
+ if (!RenameFn)
+ return;
+ }
+ }
if (unsigned IID = F->getIntrinsicID()) {
RenameFn = visitIntrinsicCall(I, IID);
if (!RenameFn)
// Check for well-known libc/libm calls. If the function is internal, it
// can't be a library call.
unsigned NameLen = F->getNameLen();
- if (!F->hasInternalLinkage() && NameLen) {
+ if (!F->hasLocalLinkage() && NameLen) {
const char *NameStr = F->getNameStart();
if (NameStr[0] == 'c' &&
((NameLen == 8 && !strcmp(NameStr, "copysign")) ||
I.getType() == I.getOperand(2)->getType()) {
SDValue LHS = getValue(I.getOperand(1));
SDValue RHS = getValue(I.getOperand(2));
- setValue(&I, DAG.getNode(ISD::FCOPYSIGN, LHS.getValueType(),
- LHS, RHS));
+ setValue(&I, DAG.getNode(ISD::FCOPYSIGN, getCurDebugLoc(),
+ LHS.getValueType(), LHS, RHS));
return;
}
} else if (NameStr[0] == 'f' &&
I.getOperand(1)->getType()->isFloatingPoint() &&
I.getType() == I.getOperand(1)->getType()) {
SDValue Tmp = getValue(I.getOperand(1));
- setValue(&I, DAG.getNode(ISD::FABS, Tmp.getValueType(), Tmp));
+ setValue(&I, DAG.getNode(ISD::FABS, getCurDebugLoc(),
+ Tmp.getValueType(), Tmp));
return;
}
- } else if (NameStr[0] == 's' &&
+ } else if (NameStr[0] == 's' &&
((NameLen == 3 && !strcmp(NameStr, "sin")) ||
(NameLen == 4 && !strcmp(NameStr, "sinf")) ||
(NameLen == 4 && !strcmp(NameStr, "sinl")))) {
I.getOperand(1)->getType()->isFloatingPoint() &&
I.getType() == I.getOperand(1)->getType()) {
SDValue Tmp = getValue(I.getOperand(1));
- setValue(&I, DAG.getNode(ISD::FSIN, Tmp.getValueType(), Tmp));
+ setValue(&I, DAG.getNode(ISD::FSIN, getCurDebugLoc(),
+ Tmp.getValueType(), Tmp));
return;
}
} else if (NameStr[0] == 'c' &&
I.getOperand(1)->getType()->isFloatingPoint() &&
I.getType() == I.getOperand(1)->getType()) {
SDValue Tmp = getValue(I.getOperand(1));
- setValue(&I, DAG.getNode(ISD::FCOS, Tmp.getValueType(), Tmp));
+ setValue(&I, DAG.getNode(ISD::FCOS, getCurDebugLoc(),
+ Tmp.getValueType(), Tmp));
return;
}
}
/// getCopyFromRegs - Emit a series of CopyFromReg nodes that copies from
-/// this value and returns the result as a ValueVT value. This uses
+/// this value and returns the result as a ValueVT value. This uses
/// Chain/Flag as the input and updates them for the output Chain/Flag.
/// If the Flag pointer is NULL, no flag is used.
-SDValue RegsForValue::getCopyFromRegs(SelectionDAG &DAG,
+SDValue RegsForValue::getCopyFromRegs(SelectionDAG &DAG, DebugLoc dl,
SDValue &Chain,
SDValue *Flag) const {
// Assemble the legal parts into the final values.
for (unsigned i = 0; i != NumRegs; ++i) {
SDValue P;
if (Flag == 0)
- P = DAG.getCopyFromReg(Chain, Regs[Part+i], RegisterVT);
+ P = DAG.getCopyFromReg(Chain, dl, Regs[Part+i], RegisterVT);
else {
- P = DAG.getCopyFromReg(Chain, Regs[Part+i], RegisterVT, *Flag);
+ P = DAG.getCopyFromReg(Chain, dl, Regs[Part+i], RegisterVT, *Flag);
*Flag = P.getValue(2);
}
Chain = P.getValue(1);
-
+
// If the source register was virtual and if we know something about it,
// add an assert node.
if (TargetRegisterInfo::isVirtualRegister(Regs[Part+i]) &&
FunctionLoweringInfo &FLI = DAG.getFunctionLoweringInfo();
if (FLI.LiveOutRegInfo.size() > SlotNo) {
FunctionLoweringInfo::LiveOutInfo &LOI = FLI.LiveOutRegInfo[SlotNo];
-
+
unsigned RegSize = RegisterVT.getSizeInBits();
unsigned NumSignBits = LOI.NumSignBits;
unsigned NumZeroBits = LOI.KnownZero.countLeadingOnes();
-
+
// FIXME: We capture more information than the dag can represent. For
// now, just use the tightest assertzext/assertsext possible.
bool isSExt = true;
else if (NumZeroBits >= RegSize-9)
isSExt = false, FromVT = MVT::i8; // ASSERT ZEXT 8
else if (NumSignBits > RegSize-16)
- isSExt = true, FromVT = MVT::i16; // ASSERT SEXT 16
+ isSExt = true, FromVT = MVT::i16; // ASSERT SEXT 16
else if (NumZeroBits >= RegSize-17)
- isSExt = false, FromVT = MVT::i16; // ASSERT ZEXT 16
+ isSExt = false, FromVT = MVT::i16; // ASSERT ZEXT 16
else if (NumSignBits > RegSize-32)
- isSExt = true, FromVT = MVT::i32; // ASSERT SEXT 32
+ isSExt = true, FromVT = MVT::i32; // ASSERT SEXT 32
else if (NumZeroBits >= RegSize-33)
- isSExt = false, FromVT = MVT::i32; // ASSERT ZEXT 32
-
+ isSExt = false, FromVT = MVT::i32; // ASSERT ZEXT 32
+
if (FromVT != MVT::Other) {
- P = DAG.getNode(isSExt ? ISD::AssertSext : ISD::AssertZext,
+ P = DAG.getNode(isSExt ? ISD::AssertSext : ISD::AssertZext, dl,
RegisterVT, P, DAG.getValueType(FromVT));
}
}
}
-
+
Parts[i] = P;
}
-
- Values[Value] = getCopyFromParts(DAG, Parts.begin(), NumRegs, RegisterVT,
- ValueVT);
+
+ Values[Value] = getCopyFromParts(DAG, dl, Parts.begin(),
+ NumRegs, RegisterVT, ValueVT);
Part += NumRegs;
Parts.clear();
}
- return DAG.getMergeValues(DAG.getVTList(&ValueVTs[0], ValueVTs.size()),
- &Values[0], ValueVTs.size());
+ return DAG.getNode(ISD::MERGE_VALUES, dl,
+ DAG.getVTList(&ValueVTs[0], ValueVTs.size()),
+ &Values[0], ValueVTs.size());
}
/// getCopyToRegs - Emit a series of CopyToReg nodes that copies the
-/// specified value into the registers specified by this object. This uses
+/// specified value into the registers specified by this object. This uses
/// Chain/Flag as the input and updates them for the output Chain/Flag.
/// If the Flag pointer is NULL, no flag is used.
-void RegsForValue::getCopyToRegs(SDValue Val, SelectionDAG &DAG,
+void RegsForValue::getCopyToRegs(SDValue Val, SelectionDAG &DAG, DebugLoc dl,
SDValue &Chain, SDValue *Flag) const {
// Get the list of the values's legal parts.
unsigned NumRegs = Regs.size();
unsigned NumParts = TLI->getNumRegisters(ValueVT);
MVT RegisterVT = RegVTs[Value];
- getCopyToParts(DAG, Val.getValue(Val.getResNo() + Value),
+ getCopyToParts(DAG, dl, Val.getValue(Val.getResNo() + Value),
&Parts[Part], NumParts, RegisterVT);
Part += NumParts;
}
for (unsigned i = 0; i != NumRegs; ++i) {
SDValue Part;
if (Flag == 0)
- Part = DAG.getCopyToReg(Chain, Regs[i], Parts[i]);
+ Part = DAG.getCopyToReg(Chain, dl, Regs[i], Parts[i]);
else {
- Part = DAG.getCopyToReg(Chain, Regs[i], Parts[i], *Flag);
+ Part = DAG.getCopyToReg(Chain, dl, Regs[i], Parts[i], *Flag);
*Flag = Part.getValue(1);
}
Chains[i] = Part.getValue(0);
}
-
+
if (NumRegs == 1 || Flag)
- // If NumRegs > 1 && Flag is used then the use of the last CopyToReg is
+ // If NumRegs > 1 && Flag is used then the use of the last CopyToReg is
// flagged to it. That is the CopyToReg nodes and the user are considered
// a single scheduling unit. If we create a TokenFactor and return it as
// chain, then the TokenFactor is both a predecessor (operand) of the
// = op c3, ..., f2
Chain = Chains[NumRegs-1];
else
- Chain = DAG.getNode(ISD::TokenFactor, MVT::Other, &Chains[0], NumRegs);
+ Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, &Chains[0], NumRegs);
}
/// AddInlineAsmOperands - Add this value to the specified inlineasm node
-/// operand list. This adds the code marker and includes the number of
+/// operand list. This adds the code marker and includes the number of
/// values added into it.
void RegsForValue::AddInlineAsmOperands(unsigned Code, SelectionDAG &DAG,
std::vector<SDValue> &Ops) const {
for (unsigned Value = 0, Reg = 0, e = ValueVTs.size(); Value != e; ++Value) {
unsigned NumRegs = TLI->getNumRegisters(ValueVTs[Value]);
MVT RegisterVT = RegVTs[Value];
- for (unsigned i = 0; i != NumRegs; ++i)
+ for (unsigned i = 0; i != NumRegs; ++i) {
+ assert(Reg < Regs.size() && "Mismatch in # registers expected");
Ops.push_back(DAG.getRegister(Regs[Reg++], RegisterVT));
+ }
}
}
-/// isAllocatableRegister - If the specified register is safe to allocate,
+/// isAllocatableRegister - If the specified register is safe to allocate,
/// i.e. it isn't a stack pointer or some other special register, return the
/// register class for the register. Otherwise, return null.
static const TargetRegisterClass *
MVT ThisVT = MVT::Other;
const TargetRegisterClass *RC = *RCI;
- // If none of the the value types for this register class are valid, we
+ // If none of the the value types for this register class are valid, we
// can't use it. For example, 64-bit reg classes on 32-bit targets.
for (TargetRegisterClass::vt_iterator I = RC->vt_begin(), E = RC->vt_end();
I != E; ++I) {
}
}
}
-
+
if (ThisVT == MVT::Other) continue;
-
+
// NOTE: This isn't ideal. In particular, this might allocate the
// frame pointer in functions that need it (due to them not being taken
// out of allocation, because a variable sized allocation hasn't been seen
}
}
return FoundRC;
-}
+}
namespace llvm {
/// AsmOperandInfo - This contains information for each constraint that we are
/// lowering.
-struct VISIBILITY_HIDDEN SDISelAsmOperandInfo :
+struct VISIBILITY_HIDDEN SDISelAsmOperandInfo :
public TargetLowering::AsmOperandInfo {
/// CallOperand - If this is the result output operand or a clobber
/// this is null, otherwise it is the incoming operand to the CallInst.
/// AssignedRegs - If this is a register or register class operand, this
/// contains the set of register corresponding to the operand.
RegsForValue AssignedRegs;
-
+
explicit SDISelAsmOperandInfo(const InlineAsm::ConstraintInfo &info)
: 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> &OutputRegs,
std::set<unsigned> &InputRegs,
const TargetRegisterInfo &TRI) const {
if (isOutReg) {
MarkRegAndAliases(AssignedRegs.Regs[i], InputRegs, TRI);
}
}
-
+
+ /// getCallOperandValMVT - Return the MVT of the Value* that this operand
+ /// corresponds to. If there is no Value* for this operand, it returns
+ /// MVT::Other.
+ MVT getCallOperandValMVT(const TargetLowering &TLI,
+ const TargetData *TD) const {
+ if (CallOperandVal == 0) return MVT::Other;
+
+ if (isa<BasicBlock>(CallOperandVal))
+ return TLI.getPointerTy();
+
+ const llvm::Type *OpTy = CallOperandVal->getType();
+
+ // If this is an indirect operand, the operand is a pointer to the
+ // accessed type.
+ if (isIndirect)
+ OpTy = cast<PointerType>(OpTy)->getElementType();
+
+ // 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);
+ switch (BitSize) {
+ default: break;
+ case 1:
+ case 8:
+ case 16:
+ case 32:
+ case 64:
+ case 128:
+ OpTy = IntegerType::get(BitSize);
+ break;
+ }
+ }
+
+ 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,
+ static void MarkRegAndAliases(unsigned Reg, std::set<unsigned> &Regs,
const TargetRegisterInfo &TRI) {
assert(TargetRegisterInfo::isPhysicalRegister(Reg) && "Isn't a physreg");
Regs.insert(Reg);
///
void SelectionDAGLowering::
GetRegistersForValue(SDISelAsmOperandInfo &OpInfo,
- std::set<unsigned> &OutputRegs,
+ std::set<unsigned> &OutputRegs,
std::set<unsigned> &InputRegs) {
// Compute whether this value requires an input register, an output register,
// or both.
switch (OpInfo.Type) {
case InlineAsm::isOutput:
isOutReg = true;
-
- // If there is an input constraint that matches this, we need to reserve
+
+ // 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;
+ isInReg = OpInfo.hasMatchingInput();
break;
case InlineAsm::isInput:
isInReg = true;
isInReg = true;
break;
}
-
-
+
+
MachineFunction &MF = DAG.getMachineFunction();
SmallVector<unsigned, 4> Regs;
-
+
// If this is a constraint for a single physreg, or a constraint for a
// register class, find it.
- std::pair<unsigned, const TargetRegisterClass*> PhysReg =
+ std::pair<unsigned, const TargetRegisterClass*> PhysReg =
TLI.getRegForInlineAsmConstraint(OpInfo.ConstraintCode,
OpInfo.ConstraintVT);
unsigned NumRegs = 1;
- if (OpInfo.ConstraintVT != MVT::Other)
+ if (OpInfo.ConstraintVT != MVT::Other) {
+ // If this is a FP input in an integer register (or visa versa) insert a bit
+ // cast of the input value. More generally, handle any case where the input
+ // value disagrees with the register class we plan to stick this in.
+ if (OpInfo.Type == InlineAsm::isInput &&
+ PhysReg.second && !PhysReg.second->hasType(OpInfo.ConstraintVT)) {
+ // Try to convert to the first MVT that the reg class contains. If the
+ // types are identical size, use a bitcast to convert (e.g. two differing
+ // vector types).
+ MVT RegVT = *PhysReg.second->vt_begin();
+ if (RegVT.getSizeInBits() == OpInfo.ConstraintVT.getSizeInBits()) {
+ OpInfo.CallOperand = DAG.getNode(ISD::BIT_CONVERT, getCurDebugLoc(),
+ RegVT, OpInfo.CallOperand);
+ OpInfo.ConstraintVT = RegVT;
+ } else if (RegVT.isInteger() && OpInfo.ConstraintVT.isFloatingPoint()) {
+ // If the input is a FP value and we want it in FP registers, do a
+ // 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 = MVT::getIntegerVT(OpInfo.ConstraintVT.getSizeInBits());
+ OpInfo.CallOperand = DAG.getNode(ISD::BIT_CONVERT, getCurDebugLoc(),
+ RegVT, OpInfo.CallOperand);
+ OpInfo.ConstraintVT = RegVT;
+ }
+ }
+
NumRegs = TLI.getNumRegisters(OpInfo.ConstraintVT);
+ }
+
MVT RegVT;
MVT ValueVT = OpInfo.ConstraintVT;
-
// If this is a constraint for a specific physical register, like {r17},
// assign it now.
if (PhysReg.first) {
if (OpInfo.ConstraintVT == MVT::Other)
ValueVT = *PhysReg.second->vt_begin();
-
+
// Get the actual register value type. This is important, because the user
// may have asked for (e.g.) the AX register in i32 type. We need to
// remember that AX is actually i16 to get the right extension.
RegVT = *PhysReg.second->vt_begin();
-
+
// This is a explicit reference to a physical register.
Regs.push_back(PhysReg.first);
if (NumRegs != 1) {
TargetRegisterClass::iterator I = PhysReg.second->begin();
for (; *I != PhysReg.first; ++I)
- assert(I != PhysReg.second->end() && "Didn't find reg!");
-
+ assert(I != PhysReg.second->end() && "Didn't find reg!");
+
// Already added the first reg.
--NumRegs; ++I;
for (; NumRegs; --NumRegs, ++I) {
OpInfo.MarkAllocatedRegs(isOutReg, isInReg, OutputRegs, InputRegs, *TRI);
return;
}
-
+
// Otherwise, if this was a reference to an LLVM register class, create vregs
// for this reference.
std::vector<unsigned> RegClassRegs;
const TargetRegisterClass *RC = PhysReg.second;
if (RC) {
- // If this is a tied register, our regalloc doesn't know how to maintain
- // the constraint. If it isn't, go ahead and create vreg
- // and let the regalloc do the right thing.
- if (!OpInfo.hasMatchingInput) {
+ // If this is a tied register, our regalloc doesn't know how to maintain
+ // the constraint, so we have to pick a register to pin the input/output to.
+ // If it isn't a matched constraint, go ahead and create vreg and let the
+ // regalloc do its thing.
+ if (!OpInfo.hasMatchingInput()) {
RegVT = *PhysReg.second->vt_begin();
if (OpInfo.ConstraintVT == MVT::Other)
ValueVT = RegVT;
MachineRegisterInfo &RegInfo = MF.getRegInfo();
for (; NumRegs; --NumRegs)
Regs.push_back(RegInfo.createVirtualRegister(PhysReg.second));
-
+
OpInfo.AssignedRegs = RegsForValue(TLI, Regs, RegVT, ValueVT);
return;
}
-
+
// Otherwise, we can't allocate it. Let the code below figure out how to
// maintain these constraints.
RegClassRegs.assign(PhysReg.second->begin(), PhysReg.second->end());
-
+
} else {
// This is a reference to a register class that doesn't directly correspond
// to an LLVM register class. Allocate NumRegs consecutive, available,
RegClassRegs = TLI.getRegClassForInlineAsmConstraint(OpInfo.ConstraintCode,
OpInfo.ConstraintVT);
}
-
+
const TargetRegisterInfo *TRI = DAG.getTarget().getRegisterInfo();
unsigned NumAllocated = 0;
for (unsigned i = 0, e = RegClassRegs.size(); i != e; ++i) {
NumAllocated = 0;
continue;
}
-
+
// Check to see if this register is allocatable (i.e. don't give out the
// stack pointer).
if (RC == 0) {
continue;
}
}
-
+
// Okay, this register is good, we can use it.
++NumAllocated;
// Mark all of the allocated registers used.
for (unsigned i = RegStart; i != RegEnd; ++i)
Regs.push_back(RegClassRegs[i]);
-
- OpInfo.AssignedRegs = RegsForValue(TLI, Regs, *RC->vt_begin(),
+
+ OpInfo.AssignedRegs = RegsForValue(TLI, Regs, *RC->vt_begin(),
OpInfo.ConstraintVT);
OpInfo.MarkAllocatedRegs(isOutReg, isInReg, OutputRegs, InputRegs, *TRI);
return;
}
}
-
+
// Otherwise, we couldn't allocate enough registers for this.
}
/// processed uses a memory 'm' constraint.
static bool
hasInlineAsmMemConstraint(std::vector<InlineAsm::ConstraintInfo> &CInfos,
- TargetLowering &TLI) {
+ const TargetLowering &TLI) {
for (unsigned i = 0, e = CInfos.size(); i != e; ++i) {
InlineAsm::ConstraintInfo &CI = CInfos[i];
for (unsigned j = 0, ee = CI.Codes.size(); j != ee; ++j) {
/// ConstraintOperands - Information about all of the constraints.
std::vector<SDISelAsmOperandInfo> ConstraintOperands;
-
+
SDValue Chain = getRoot();
SDValue Flag;
-
+
std::set<unsigned> OutputRegs, InputRegs;
// Do a prepass over the constraints, canonicalizing them, and building up the
ConstraintInfos = IA->ParseConstraints();
bool hasMemory = hasInlineAsmMemConstraint(ConstraintInfos, TLI);
-
+
unsigned ArgNo = 0; // ArgNo - The argument of the CallInst.
unsigned ResNo = 0; // ResNo - The result number of the next output.
for (unsigned i = 0, e = ConstraintInfos.size(); i != e; ++i) {
ConstraintOperands.push_back(SDISelAsmOperandInfo(ConstraintInfos[i]));
SDISelAsmOperandInfo &OpInfo = ConstraintOperands.back();
-
+
MVT OpVT = MVT::Other;
// Compute the value type for each operand.
OpInfo.CallOperandVal = CS.getArgument(ArgNo++);
break;
}
+
// The return value of the call is this value. As such, there is no
// corresponding argument.
assert(CS.getType() != Type::VoidTy && "Bad inline asm!");
// If this is an input or an indirect output, process the call argument.
// BasicBlocks are labels, currently appearing only in asm's.
if (OpInfo.CallOperandVal) {
- if (BasicBlock *BB = dyn_cast<BasicBlock>(OpInfo.CallOperandVal))
+ if (BasicBlock *BB = dyn_cast<BasicBlock>(OpInfo.CallOperandVal)) {
OpInfo.CallOperand = DAG.getBasicBlock(FuncInfo.MBBMap[BB]);
- else {
+ } else {
OpInfo.CallOperand = getValue(OpInfo.CallOperandVal);
- const Type *OpTy = OpInfo.CallOperandVal->getType();
- // If this is an indirect operand, the operand is a pointer to the
- // accessed type.
- if (OpInfo.isIndirect)
- OpTy = cast<PointerType>(OpTy)->getElementType();
-
- // 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);
- switch (BitSize) {
- default: break;
- case 1:
- case 8:
- case 16:
- case 32:
- case 64:
- OpTy = IntegerType::get(BitSize);
- break;
- }
- }
-
- OpVT = TLI.getValueType(OpTy, true);
}
+
+ OpVT = OpInfo.getCallOperandValMVT(TLI, TD);
}
-
+
OpInfo.ConstraintVT = OpVT;
-
+ }
+
+ // Second pass over the constraints: compute which constraint option to use
+ // and assign registers to constraints that want a specific physreg.
+ for (unsigned i = 0, e = ConstraintInfos.size(); i != e; ++i) {
+ SDISelAsmOperandInfo &OpInfo = ConstraintOperands[i];
+
+ // If this is an output operand with a matching input operand, look up the
+ // matching input. If their types mismatch, e.g. one is an integer, the
+ // other is floating point, or their sizes are different, flag it as an
+ // error.
+ if (OpInfo.hasMatchingInput()) {
+ SDISelAsmOperandInfo &Input = ConstraintOperands[OpInfo.MatchingInput];
+ if (OpInfo.ConstraintVT != Input.ConstraintVT) {
+ if ((OpInfo.ConstraintVT.isInteger() !=
+ Input.ConstraintVT.isInteger()) ||
+ (OpInfo.ConstraintVT.getSizeInBits() !=
+ Input.ConstraintVT.getSizeInBits())) {
+ cerr << "Unsupported asm: input constraint with a matching output "
+ << "constraint of incompatible type!\n";
+ exit(1);
+ }
+ Input.ConstraintVT = OpInfo.ConstraintVT;
+ }
+ }
+
// Compute the constraint code and ConstraintType to use.
TLI.ComputeConstraintToUse(OpInfo, OpInfo.CallOperand, hasMemory, &DAG);
!OpInfo.isIndirect) {
assert(OpInfo.Type == InlineAsm::isInput &&
"Can only indirectify direct input operands!");
-
+
// Memory operands really want the address of the value. If we don't have
// an indirect input, put it in the constpool if we can, otherwise spill
// it to a stack slot.
-
+
// If the operand is a float, integer, or vector constant, spill to a
// constant pool entry to get its address.
Value *OpVal = OpInfo.CallOperandVal;
// Otherwise, create a stack slot and emit a store to it before the
// asm.
const Type *Ty = OpVal->getType();
- uint64_t TySize = TLI.getTargetData()->getABITypeSize(Ty);
+ uint64_t TySize = TLI.getTargetData()->getTypePaddedSize(Ty);
unsigned Align = TLI.getTargetData()->getPrefTypeAlignment(Ty);
MachineFunction &MF = DAG.getMachineFunction();
int SSFI = MF.getFrameInfo()->CreateStackObject(TySize, Align);
SDValue StackSlot = DAG.getFrameIndex(SSFI, TLI.getPointerTy());
- Chain = DAG.getStore(Chain, OpInfo.CallOperand, StackSlot, NULL, 0);
+ Chain = DAG.getStore(Chain, getCurDebugLoc(),
+ OpInfo.CallOperand, StackSlot, NULL, 0);
OpInfo.CallOperand = StackSlot;
}
-
+
// There is no longer a Value* corresponding to this operand.
OpInfo.CallOperandVal = 0;
// It is now an indirect operand.
OpInfo.isIndirect = true;
}
-
+
// If this constraint is for a specific register, allocate it before
// anything else.
if (OpInfo.ConstraintType == TargetLowering::C_Register)
GetRegistersForValue(OpInfo, OutputRegs, InputRegs);
}
ConstraintInfos.clear();
-
-
+
+
// Second pass - Loop over all of the operands, assigning virtual or physregs
- // to registerclass operands.
+ // to register class operands.
for (unsigned i = 0, e = ConstraintOperands.size(); i != e; ++i) {
SDISelAsmOperandInfo &OpInfo = ConstraintOperands[i];
-
+
// C_Register operands have already been allocated, Other/Memory don't need
// to be.
if (OpInfo.ConstraintType == TargetLowering::C_RegisterClass)
GetRegistersForValue(OpInfo, OutputRegs, InputRegs);
- }
-
+ }
+
// AsmNodeOperands - The operands for the ISD::INLINEASM node.
std::vector<SDValue> AsmNodeOperands;
AsmNodeOperands.push_back(SDValue()); // reserve space for input chain
AsmNodeOperands.push_back(
DAG.getTargetExternalSymbol(IA->getAsmString().c_str(), MVT::Other));
-
-
+
+
// Loop over all of the inputs, copying the operand values into the
// appropriate registers and processing the output regs.
RegsForValue RetValRegs;
-
+
// IndirectStoresToEmit - The set of stores to emit after the inline asm node.
std::vector<std::pair<RegsForValue, Value*> > IndirectStoresToEmit;
-
+
for (unsigned i = 0, e = ConstraintOperands.size(); i != e; ++i) {
SDISelAsmOperandInfo &OpInfo = ConstraintOperands[i];
// Concatenate this output onto the outputs list.
RetValRegs.append(OpInfo.AssignedRegs);
}
-
+
// Add information to the INLINEASM node to know that this register is
// set.
OpInfo.AssignedRegs.AddInlineAsmOperands(OpInfo.isEarlyClobber ?
}
case InlineAsm::isInput: {
SDValue InOperandVal = OpInfo.CallOperand;
-
- if (isdigit(OpInfo.ConstraintCode[0])) { // Matching constraint?
+
+ if (OpInfo.isMatchingInputConstraint()) { // Matching constraint?
// If this is required to match an output register we have already set,
// just use its register.
- unsigned OperandNo = atoi(OpInfo.ConstraintCode.c_str());
-
+ unsigned OperandNo = OpInfo.getMatchedOperand();
+
// Scan until we find the definition we already emitted of this operand.
// When we find it, create a RegsForValue operand.
unsigned CurOp = 2; // The first operand.
for (; OperandNo; --OperandNo) {
// Advance to the next operand.
- unsigned NumOps =
+ unsigned NumOps =
cast<ConstantSDNode>(AsmNodeOperands[CurOp])->getZExtValue();
assert(((NumOps & 7) == 2 /*REGDEF*/ ||
(NumOps & 7) == 6 /*EARLYCLOBBER REGDEF*/ ||
CurOp += (NumOps>>3)+1;
}
- unsigned NumOps =
+ unsigned NumOps =
cast<ConstantSDNode>(AsmNodeOperands[CurOp])->getZExtValue();
- if ((NumOps & 7) == 2 /*REGDEF*/
+ if ((NumOps & 7) == 2 /*REGDEF*/
|| (NumOps & 7) == 6 /* EARLYCLOBBER REGDEF */) {
// Add NumOps>>3 registers to MatchedRegs.
RegsForValue MatchedRegs;
cast<RegisterSDNode>(AsmNodeOperands[++CurOp])->getReg();
MatchedRegs.Regs.push_back(Reg);
}
-
- // Use the produced MatchedRegs object to
- MatchedRegs.getCopyToRegs(InOperandVal, DAG, Chain, &Flag);
+
+ // Use the produced MatchedRegs object to
+ MatchedRegs.getCopyToRegs(InOperandVal, DAG, getCurDebugLoc(),
+ Chain, &Flag);
MatchedRegs.AddInlineAsmOperands(1 /*REGUSE*/, DAG, AsmNodeOperands);
break;
} else {
assert(((NumOps & 7) == 4) && "Unknown matching constraint!");
- assert((NumOps >> 3) == 1 && "Unexpected number of operands");
+ assert((NumOps >> 3) == 1 && "Unexpected number of operands");
// Add information to the INLINEASM node to know about this input.
AsmNodeOperands.push_back(DAG.getTargetConstant(NumOps,
TLI.getPointerTy()));
break;
}
}
-
+
if (OpInfo.ConstraintType == TargetLowering::C_Other) {
- assert(!OpInfo.isIndirect &&
+ assert(!OpInfo.isIndirect &&
"Don't know how to handle indirect other inputs yet!");
-
+
std::vector<SDValue> Ops;
TLI.LowerAsmOperandForConstraint(InOperandVal, OpInfo.ConstraintCode[0],
hasMemory, Ops, DAG);
<< OpInfo.ConstraintCode << "'!\n";
exit(1);
}
-
+
// Add information to the INLINEASM node to know about this input.
unsigned ResOpType = 3 /*IMM*/ | (Ops.size() << 3);
- AsmNodeOperands.push_back(DAG.getTargetConstant(ResOpType,
+ AsmNodeOperands.push_back(DAG.getTargetConstant(ResOpType,
TLI.getPointerTy()));
AsmNodeOperands.insert(AsmNodeOperands.end(), Ops.begin(), Ops.end());
break;
assert(OpInfo.isIndirect && "Operand must be indirect to be a mem!");
assert(InOperandVal.getValueType() == TLI.getPointerTy() &&
"Memory operands expect pointer values");
-
+
// Add information to the INLINEASM node to know about this input.
unsigned ResOpType = 4/*MEM*/ | (1<<3);
AsmNodeOperands.push_back(DAG.getTargetConstant(ResOpType,
AsmNodeOperands.push_back(InOperandVal);
break;
}
-
+
assert((OpInfo.ConstraintType == TargetLowering::C_RegisterClass ||
OpInfo.ConstraintType == TargetLowering::C_Register) &&
"Unknown constraint type!");
- assert(!OpInfo.isIndirect &&
+ assert(!OpInfo.isIndirect &&
"Don't know how to handle indirect register inputs yet!");
// Copy the input into the appropriate registers.
exit(1);
}
- OpInfo.AssignedRegs.getCopyToRegs(InOperandVal, DAG, Chain, &Flag);
-
+ OpInfo.AssignedRegs.getCopyToRegs(InOperandVal, DAG, getCurDebugLoc(),
+ Chain, &Flag);
+
OpInfo.AssignedRegs.AddInlineAsmOperands(1/*REGUSE*/,
DAG, AsmNodeOperands);
break;
}
}
}
-
+
// Finish up input operands.
AsmNodeOperands[0] = Chain;
if (Flag.getNode()) AsmNodeOperands.push_back(Flag);
-
- Chain = DAG.getNode(ISD::INLINEASM,
+
+ Chain = DAG.getNode(ISD::INLINEASM, getCurDebugLoc(),
DAG.getNodeValueTypes(MVT::Other, MVT::Flag), 2,
&AsmNodeOperands[0], AsmNodeOperands.size());
Flag = Chain.getValue(1);
// If this asm returns a register value, copy the result from that register
// and set it as the value of the call.
if (!RetValRegs.Regs.empty()) {
- SDValue Val = RetValRegs.getCopyFromRegs(DAG, Chain, &Flag);
-
- // If any of the results of the inline asm is a vector, it may have the
- // wrong width/num elts. This can happen for register classes that can
- // contain multiple different value types. The preg or vreg allocated may
- // not have the same VT as was expected. Convert it to the right type with
- // bit_convert.
- if (const StructType *ResSTy = dyn_cast<StructType>(CS.getType())) {
- for (unsigned i = 0, e = ResSTy->getNumElements(); i != e; ++i) {
- if (Val.getNode()->getValueType(i).isVector())
- Val = DAG.getNode(ISD::BIT_CONVERT,
- TLI.getValueType(ResSTy->getElementType(i)), Val);
+ SDValue Val = RetValRegs.getCopyFromRegs(DAG, getCurDebugLoc(),
+ Chain, &Flag);
+
+ // FIXME: Why don't we do this for inline asms with MRVs?
+ if (CS.getType()->isSingleValueType() && CS.getType()->isSized()) {
+ MVT ResultType = TLI.getValueType(CS.getType());
+
+ // If any of the results of the inline asm is a vector, it may have the
+ // wrong width/num elts. This can happen for register classes that can
+ // contain multiple different value types. The preg or vreg allocated may
+ // not have the same VT as was expected. Convert it to the right type
+ // with bit_convert.
+ if (ResultType != Val.getValueType() && Val.getValueType().isVector()) {
+ Val = DAG.getNode(ISD::BIT_CONVERT, getCurDebugLoc(),
+ ResultType, Val);
+
+ } else if (ResultType != Val.getValueType() &&
+ ResultType.isInteger() && Val.getValueType().isInteger()) {
+ // If a result value was tied to an input value, the computed result may
+ // have a wider width than the expected result. Extract the relevant
+ // portion.
+ Val = DAG.getNode(ISD::TRUNCATE, getCurDebugLoc(), ResultType, Val);
}
- } else {
- if (Val.getValueType().isVector())
- Val = DAG.getNode(ISD::BIT_CONVERT, TLI.getValueType(CS.getType()),
- Val);
+
+ assert(ResultType == Val.getValueType() && "Asm result value mismatch!");
}
setValue(CS.getInstruction(), Val);
}
-
+
std::vector<std::pair<SDValue, Value*> > StoresToEmit;
-
+
// Process indirect outputs, first output all of the flagged copies out of
// physregs.
for (unsigned i = 0, e = IndirectStoresToEmit.size(); i != e; ++i) {
RegsForValue &OutRegs = IndirectStoresToEmit[i].first;
Value *Ptr = IndirectStoresToEmit[i].second;
- SDValue OutVal = OutRegs.getCopyFromRegs(DAG, Chain, &Flag);
+ SDValue OutVal = OutRegs.getCopyFromRegs(DAG, getCurDebugLoc(),
+ Chain, &Flag);
StoresToEmit.push_back(std::make_pair(OutVal, Ptr));
}
-
+
// Emit the non-flagged stores from the physregs.
SmallVector<SDValue, 8> OutChains;
for (unsigned i = 0, e = StoresToEmit.size(); i != e; ++i)
- OutChains.push_back(DAG.getStore(Chain, StoresToEmit[i].first,
+ OutChains.push_back(DAG.getStore(Chain, getCurDebugLoc(),
+ StoresToEmit[i].first,
getValue(StoresToEmit[i].second),
StoresToEmit[i].second, 0));
if (!OutChains.empty())
- Chain = DAG.getNode(ISD::TokenFactor, MVT::Other,
+ Chain = DAG.getNode(ISD::TokenFactor, getCurDebugLoc(), MVT::Other,
&OutChains[0], OutChains.size());
DAG.setRoot(Chain);
}
MVT IntPtr = TLI.getPointerTy();
if (IntPtr.bitsLT(Src.getValueType()))
- Src = DAG.getNode(ISD::TRUNCATE, IntPtr, Src);
+ Src = DAG.getNode(ISD::TRUNCATE, getCurDebugLoc(), IntPtr, Src);
else if (IntPtr.bitsGT(Src.getValueType()))
- Src = DAG.getNode(ISD::ZERO_EXTEND, IntPtr, Src);
+ Src = DAG.getNode(ISD::ZERO_EXTEND, getCurDebugLoc(), IntPtr, Src);
// Scale the source by the type size.
- uint64_t ElementSize = TD->getABITypeSize(I.getType()->getElementType());
- Src = DAG.getNode(ISD::MUL, Src.getValueType(),
+ uint64_t ElementSize = TD->getTypePaddedSize(I.getType()->getElementType());
+ Src = DAG.getNode(ISD::MUL, getCurDebugLoc(), Src.getValueType(),
Src, DAG.getIntPtrConstant(ElementSize));
TargetLowering::ArgListTy Args;
Args.push_back(Entry);
std::pair<SDValue,SDValue> Result =
- TLI.LowerCallTo(getRoot(), I.getType(), false, false, false, CallingConv::C,
- PerformTailCallOpt, DAG.getExternalSymbol("malloc", IntPtr),
- Args, DAG);
+ TLI.LowerCallTo(getRoot(), I.getType(), false, false, false, false,
+ CallingConv::C, PerformTailCallOpt,
+ DAG.getExternalSymbol("malloc", IntPtr),
+ Args, DAG, getCurDebugLoc());
setValue(&I, Result.first); // Pointers always fit in registers
DAG.setRoot(Result.second);
}
Args.push_back(Entry);
MVT IntPtr = TLI.getPointerTy();
std::pair<SDValue,SDValue> Result =
- TLI.LowerCallTo(getRoot(), Type::VoidTy, false, false, false,
+ TLI.LowerCallTo(getRoot(), Type::VoidTy, false, false, false, false,
CallingConv::C, PerformTailCallOpt,
- DAG.getExternalSymbol("free", IntPtr), Args, DAG);
+ DAG.getExternalSymbol("free", IntPtr), Args, DAG,
+ getCurDebugLoc());
DAG.setRoot(Result.second);
}
void SelectionDAGLowering::visitVAStart(CallInst &I) {
- DAG.setRoot(DAG.getNode(ISD::VASTART, MVT::Other, getRoot(),
- getValue(I.getOperand(1)),
+ DAG.setRoot(DAG.getNode(ISD::VASTART, getCurDebugLoc(),
+ MVT::Other, getRoot(),
+ getValue(I.getOperand(1)),
DAG.getSrcValue(I.getOperand(1))));
}
void SelectionDAGLowering::visitVAArg(VAArgInst &I) {
- SDValue V = DAG.getVAArg(TLI.getValueType(I.getType()), getRoot(),
- getValue(I.getOperand(0)),
- DAG.getSrcValue(I.getOperand(0)));
+ SDValue V = DAG.getVAArg(TLI.getValueType(I.getType()), getCurDebugLoc(),
+ getRoot(), getValue(I.getOperand(0)),
+ DAG.getSrcValue(I.getOperand(0)));
setValue(&I, V);
DAG.setRoot(V.getValue(1));
}
void SelectionDAGLowering::visitVAEnd(CallInst &I) {
- DAG.setRoot(DAG.getNode(ISD::VAEND, MVT::Other, getRoot(),
- getValue(I.getOperand(1)),
+ DAG.setRoot(DAG.getNode(ISD::VAEND, getCurDebugLoc(),
+ MVT::Other, getRoot(),
+ getValue(I.getOperand(1)),
DAG.getSrcValue(I.getOperand(1))));
}
void SelectionDAGLowering::visitVACopy(CallInst &I) {
- DAG.setRoot(DAG.getNode(ISD::VACOPY, MVT::Other, getRoot(),
- getValue(I.getOperand(1)),
+ DAG.setRoot(DAG.getNode(ISD::VACOPY, getCurDebugLoc(),
+ MVT::Other, getRoot(),
+ getValue(I.getOperand(1)),
getValue(I.getOperand(2)),
DAG.getSrcValue(I.getOperand(1)),
DAG.getSrcValue(I.getOperand(2))));
/// TargetLowering::LowerArguments - This is the default LowerArguments
/// implementation, which just inserts a FORMAL_ARGUMENTS node. FIXME: When all
-/// targets are migrated to using FORMAL_ARGUMENTS, this hook should be
+/// targets are migrated to using FORMAL_ARGUMENTS, this hook should be
/// integrated into SDISel.
void TargetLowering::LowerArguments(Function &F, SelectionDAG &DAG,
- SmallVectorImpl<SDValue> &ArgValues) {
+ SmallVectorImpl<SDValue> &ArgValues,
+ DebugLoc dl) {
// Add CC# and isVararg as operands to the FORMAL_ARGUMENTS node.
SmallVector<SDValue, 3+16> Ops;
Ops.push_back(DAG.getRoot());
unsigned OriginalAlignment =
getTargetData()->getABITypeAlignment(ArgTy);
- if (F.paramHasAttr(j, ParamAttr::ZExt))
+ if (F.paramHasAttr(j, Attribute::ZExt))
Flags.setZExt();
- if (F.paramHasAttr(j, ParamAttr::SExt))
+ if (F.paramHasAttr(j, Attribute::SExt))
Flags.setSExt();
- if (F.paramHasAttr(j, ParamAttr::InReg))
+ if (F.paramHasAttr(j, Attribute::InReg))
Flags.setInReg();
- if (F.paramHasAttr(j, ParamAttr::StructRet))
+ if (F.paramHasAttr(j, Attribute::StructRet))
Flags.setSRet();
- if (F.paramHasAttr(j, ParamAttr::ByVal)) {
+ if (F.paramHasAttr(j, Attribute::ByVal)) {
Flags.setByVal();
const PointerType *Ty = cast<PointerType>(I->getType());
const Type *ElementTy = Ty->getElementType();
unsigned FrameAlign = getByValTypeAlignment(ElementTy);
- unsigned FrameSize = getTargetData()->getABITypeSize(ElementTy);
+ unsigned FrameSize = getTargetData()->getTypePaddedSize(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.
if (F.getParamAlignment(j))
Flags.setByValAlign(FrameAlign);
Flags.setByValSize(FrameSize);
}
- if (F.paramHasAttr(j, ParamAttr::Nest))
+ if (F.paramHasAttr(j, Attribute::Nest))
Flags.setNest();
Flags.setOrigAlign(OriginalAlignment);
}
RetVals.push_back(MVT::Other);
-
+
// Create the node.
- SDNode *Result = DAG.getNode(ISD::FORMAL_ARGUMENTS,
+ SDNode *Result = DAG.getNode(ISD::FORMAL_ARGUMENTS, dl,
DAG.getVTList(&RetVals[0], RetVals.size()),
&Ops[0], Ops.size()).getNode();
-
+
// Prelower FORMAL_ARGUMENTS. This isn't required for functionality, but
// allows exposing the loads that may be part of the argument access to the
// first DAGCombiner pass.
SDValue TmpRes = LowerOperation(SDValue(Result, 0), DAG);
-
+
// The number of results should match up, except that the lowered one may have
// an extra flag result.
assert((Result->getNumValues() == TmpRes.getNode()->getNumValues() ||
}
Result = TmpRes.getNode();
-
+
unsigned NumArgRegs = Result->getNumValues() - 1;
DAG.setRoot(SDValue(Result, NumArgRegs));
// Set up the return result vector.
unsigned i = 0;
unsigned Idx = 1;
- for (Function::arg_iterator I = F.arg_begin(), E = F.arg_end(); I != E;
+ for (Function::arg_iterator I = F.arg_begin(), E = F.arg_end(); I != E;
++I, ++Idx) {
SmallVector<MVT, 4> ValueVTs;
ComputeValueVTs(*this, I->getType(), ValueVTs);
Parts[j] = SDValue(Result, i++);
ISD::NodeType AssertOp = ISD::DELETED_NODE;
- if (F.paramHasAttr(Idx, ParamAttr::SExt))
+ if (F.paramHasAttr(Idx, Attribute::SExt))
AssertOp = ISD::AssertSext;
- else if (F.paramHasAttr(Idx, ParamAttr::ZExt))
+ else if (F.paramHasAttr(Idx, Attribute::ZExt))
AssertOp = ISD::AssertZext;
- ArgValues.push_back(getCopyFromParts(DAG, &Parts[0], NumParts, PartVT, VT,
- AssertOp));
+ ArgValues.push_back(getCopyFromParts(DAG, dl, &Parts[0], NumParts,
+ PartVT, VT, AssertOp));
}
}
assert(i == NumArgRegs && "Argument register count mismatch!");
std::pair<SDValue, SDValue>
TargetLowering::LowerCallTo(SDValue Chain, const Type *RetTy,
bool RetSExt, bool RetZExt, bool isVarArg,
+ bool isInreg,
unsigned CallingConv, bool isTailCall,
SDValue Callee,
- ArgListTy &Args, SelectionDAG &DAG) {
+ ArgListTy &Args, SelectionDAG &DAG, DebugLoc dl) {
assert((!isTailCall || PerformTailCallOpt) &&
"isTailCall set when tail-call optimizations are disabled!");
SmallVector<SDValue, 32> Ops;
Ops.push_back(Chain); // Op#0 - Chain
Ops.push_back(Callee);
-
+
// Handle all of the outgoing arguments.
for (unsigned i = 0, e = Args.size(); i != e; ++i) {
SmallVector<MVT, 4> ValueVTs;
Value != NumValues; ++Value) {
MVT VT = ValueVTs[Value];
const Type *ArgTy = VT.getTypeForMVT();
- SDValue Op = SDValue(Args[i].Node.getNode(), Args[i].Node.getResNo() + Value);
+ SDValue Op = SDValue(Args[i].Node.getNode(),
+ Args[i].Node.getResNo() + Value);
ISD::ArgFlagsTy Flags;
unsigned OriginalAlignment =
getTargetData()->getABITypeAlignment(ArgTy);
const PointerType *Ty = cast<PointerType>(Args[i].Ty);
const Type *ElementTy = Ty->getElementType();
unsigned FrameAlign = getByValTypeAlignment(ElementTy);
- unsigned FrameSize = getTargetData()->getABITypeSize(ElementTy);
+ unsigned FrameSize = getTargetData()->getTypePaddedSize(ElementTy);
// For ByVal, alignment should come from FE. BE will guess if this
// info is not there but there are cases it cannot get right.
if (Args[i].Alignment)
else if (Args[i].isZExt)
ExtendKind = ISD::ZERO_EXTEND;
- getCopyToParts(DAG, Op, &Parts[0], NumParts, PartVT, ExtendKind);
+ getCopyToParts(DAG, dl, Op, &Parts[0], NumParts, PartVT, ExtendKind);
for (unsigned i = 0; i != NumParts; ++i) {
// if it isn't first piece, alignment must be 1
}
}
}
-
+
// Figure out the result value types. We start by making a list of
// the potentially illegal return value types.
SmallVector<MVT, 4> LoweredRetTys;
for (unsigned i = 0; i != NumRegs; ++i)
LoweredRetTys.push_back(RegisterVT);
}
-
+
LoweredRetTys.push_back(MVT::Other); // Always has a chain.
-
+
// Create the CALL node.
- SDValue Res = DAG.getCall(CallingConv, isVarArg, isTailCall,
+ SDValue Res = DAG.getCall(CallingConv, dl,
+ isVarArg, isTailCall, isInreg,
DAG.getVTList(&LoweredRetTys[0],
LoweredRetTys.size()),
- &Ops[0], Ops.size());
+ &Ops[0], Ops.size()
+ );
Chain = Res.getValue(LoweredRetTys.size() - 1);
// Gather up the call result into a single value.
- if (RetTy != Type::VoidTy) {
+ if (RetTy != Type::VoidTy && !RetTys.empty()) {
ISD::NodeType AssertOp = ISD::DELETED_NODE;
if (RetSExt)
for (; RegNo != RegNoEnd; ++RegNo)
Results.push_back(Res.getValue(RegNo));
SDValue ReturnValue =
- getCopyFromParts(DAG, &Results[0], NumRegs, RegisterVT, VT,
+ getCopyFromParts(DAG, dl, &Results[0], NumRegs, RegisterVT, VT,
AssertOp);
ReturnValues.push_back(ReturnValue);
}
- Res = DAG.getMergeValues(DAG.getVTList(&RetTys[0], RetTys.size()),
- &ReturnValues[0], ReturnValues.size());
+ Res = DAG.getNode(ISD::MERGE_VALUES, dl,
+ DAG.getVTList(&RetTys[0], RetTys.size()),
+ &ReturnValues[0], ReturnValues.size());
}
return std::make_pair(Res, Chain);
}
+void TargetLowering::LowerOperationWrapper(SDNode *N,
+ SmallVectorImpl<SDValue> &Results,
+ SelectionDAG &DAG) {
+ SDValue Res = LowerOperation(SDValue(N, 0), DAG);
+ if (Res.getNode())
+ Results.push_back(Res);
+}
+
SDValue TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) {
assert(0 && "LowerOperation not implemented for this target!");
abort();
RegsForValue RFV(TLI, Reg, V->getType());
SDValue Chain = DAG.getEntryNode();
- RFV.getCopyToRegs(Op, DAG, Chain, 0);
+ RFV.getCopyToRegs(Op, DAG, getCurDebugLoc(), Chain, 0);
PendingExports.push_back(Chain);
}
Function &F = *LLVMBB->getParent();
SDValue OldRoot = SDL->DAG.getRoot();
SmallVector<SDValue, 16> Args;
- TLI.LowerArguments(F, SDL->DAG, Args);
+ TLI.LowerArguments(F, SDL->DAG, Args, SDL->getCurDebugLoc());
unsigned a = 0;
for (Function::arg_iterator AI = F.arg_begin(), E = F.arg_end();
BasicBlock *SuccBB = TI->getSuccessor(succ);
if (!isa<PHINode>(SuccBB->begin())) continue;
MachineBasicBlock *SuccMBB = FuncInfo->MBBMap[SuccBB];
-
+
// If this terminator has multiple identical successors (common for
// switches), only handle each succ once.
if (!SuccsHandled.insert(SuccMBB)) continue;
-
+
MachineBasicBlock::iterator MBBI = SuccMBB->begin();
PHINode *PN;
BasicBlock *SuccBB = TI->getSuccessor(succ);
if (!isa<PHINode>(SuccBB->begin())) continue;
MachineBasicBlock *SuccMBB = FuncInfo->MBBMap[SuccBB];
-
+
// If this terminator has multiple identical successors (common for
// switches), only handle each succ once.
if (!SuccsHandled.insert(SuccMBB)) continue;
-
+
MachineBasicBlock::iterator MBBI = SuccMBB->begin();
PHINode *PN;