//===----------------------------------------------------------------------===//
#include "llvm/Target/TargetLowering.h"
-#include "llvm/Target/TargetAsmInfo.h"
+#include "llvm/MC/MCAsmInfo.h"
+#include "llvm/MC/MCExpr.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Target/TargetLoweringObjectFile.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/GlobalVariable.h"
#include "llvm/DerivedTypes.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineJumpTableInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/SelectionDAG.h"
-#include "llvm/ADT/StringExtras.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MathExtras.h"
Names[RTLIB::SRA_I32] = "__ashrsi3";
Names[RTLIB::SRA_I64] = "__ashrdi3";
Names[RTLIB::SRA_I128] = "__ashrti3";
+ Names[RTLIB::MUL_I8] = "__mulqi3";
Names[RTLIB::MUL_I16] = "__mulhi3";
Names[RTLIB::MUL_I32] = "__mulsi3";
Names[RTLIB::MUL_I64] = "__muldi3";
Names[RTLIB::MUL_I128] = "__multi3";
+ Names[RTLIB::SDIV_I8] = "__divqi3";
Names[RTLIB::SDIV_I16] = "__divhi3";
Names[RTLIB::SDIV_I32] = "__divsi3";
Names[RTLIB::SDIV_I64] = "__divdi3";
Names[RTLIB::SDIV_I128] = "__divti3";
+ Names[RTLIB::UDIV_I8] = "__udivqi3";
Names[RTLIB::UDIV_I16] = "__udivhi3";
Names[RTLIB::UDIV_I32] = "__udivsi3";
Names[RTLIB::UDIV_I64] = "__udivdi3";
Names[RTLIB::UDIV_I128] = "__udivti3";
+ Names[RTLIB::SREM_I8] = "__modqi3";
Names[RTLIB::SREM_I16] = "__modhi3";
Names[RTLIB::SREM_I32] = "__modsi3";
Names[RTLIB::SREM_I64] = "__moddi3";
Names[RTLIB::SREM_I128] = "__modti3";
+ Names[RTLIB::UREM_I8] = "__umodqi3";
Names[RTLIB::UREM_I16] = "__umodhi3";
Names[RTLIB::UREM_I32] = "__umodsi3";
Names[RTLIB::UREM_I64] = "__umoddi3";
Names[RTLIB::FLOOR_F64] = "floor";
Names[RTLIB::FLOOR_F80] = "floorl";
Names[RTLIB::FLOOR_PPCF128] = "floorl";
+ Names[RTLIB::COPYSIGN_F32] = "copysignf";
+ Names[RTLIB::COPYSIGN_F64] = "copysign";
+ Names[RTLIB::COPYSIGN_F80] = "copysignl";
+ Names[RTLIB::COPYSIGN_PPCF128] = "copysignl";
Names[RTLIB::FPEXT_F32_F64] = "__extendsfdf2";
+ Names[RTLIB::FPEXT_F16_F32] = "__gnu_h2f_ieee";
+ Names[RTLIB::FPROUND_F32_F16] = "__gnu_f2h_ieee";
Names[RTLIB::FPROUND_F64_F32] = "__truncdfsf2";
Names[RTLIB::FPROUND_F80_F32] = "__truncxfsf2";
Names[RTLIB::FPROUND_PPCF128_F32] = "__trunctfsf2";
Names[RTLIB::UNWIND_RESUME] = "_Unwind_Resume";
}
+/// InitLibcallCallingConvs - Set default libcall CallingConvs.
+///
+static void InitLibcallCallingConvs(CallingConv::ID *CCs) {
+ for (int i = 0; i < RTLIB::UNKNOWN_LIBCALL; ++i) {
+ CCs[i] = CallingConv::C;
+ }
+}
+
/// getFPEXT - Return the FPEXT_*_* value for the given types, or
/// UNKNOWN_LIBCALL if there is none.
-RTLIB::Libcall RTLIB::getFPEXT(MVT OpVT, MVT RetVT) {
+RTLIB::Libcall RTLIB::getFPEXT(EVT OpVT, EVT RetVT) {
if (OpVT == MVT::f32) {
if (RetVT == MVT::f64)
return FPEXT_F32_F64;
}
+
return UNKNOWN_LIBCALL;
}
/// getFPROUND - Return the FPROUND_*_* value for the given types, or
/// UNKNOWN_LIBCALL if there is none.
-RTLIB::Libcall RTLIB::getFPROUND(MVT OpVT, MVT RetVT) {
+RTLIB::Libcall RTLIB::getFPROUND(EVT OpVT, EVT RetVT) {
if (RetVT == MVT::f32) {
if (OpVT == MVT::f64)
return FPROUND_F64_F32;
if (OpVT == MVT::ppcf128)
return FPROUND_PPCF128_F64;
}
+
return UNKNOWN_LIBCALL;
}
/// getFPTOSINT - Return the FPTOSINT_*_* value for the given types, or
/// UNKNOWN_LIBCALL if there is none.
-RTLIB::Libcall RTLIB::getFPTOSINT(MVT OpVT, MVT RetVT) {
+RTLIB::Libcall RTLIB::getFPTOSINT(EVT OpVT, EVT RetVT) {
if (OpVT == MVT::f32) {
if (RetVT == MVT::i8)
return FPTOSINT_F32_I8;
/// getFPTOUINT - Return the FPTOUINT_*_* value for the given types, or
/// UNKNOWN_LIBCALL if there is none.
-RTLIB::Libcall RTLIB::getFPTOUINT(MVT OpVT, MVT RetVT) {
+RTLIB::Libcall RTLIB::getFPTOUINT(EVT OpVT, EVT RetVT) {
if (OpVT == MVT::f32) {
if (RetVT == MVT::i8)
return FPTOUINT_F32_I8;
/// getSINTTOFP - Return the SINTTOFP_*_* value for the given types, or
/// UNKNOWN_LIBCALL if there is none.
-RTLIB::Libcall RTLIB::getSINTTOFP(MVT OpVT, MVT RetVT) {
+RTLIB::Libcall RTLIB::getSINTTOFP(EVT OpVT, EVT RetVT) {
if (OpVT == MVT::i32) {
if (RetVT == MVT::f32)
return SINTTOFP_I32_F32;
/// getUINTTOFP - Return the UINTTOFP_*_* value for the given types, or
/// UNKNOWN_LIBCALL if there is none.
-RTLIB::Libcall RTLIB::getUINTTOFP(MVT OpVT, MVT RetVT) {
+RTLIB::Libcall RTLIB::getUINTTOFP(EVT OpVT, EVT RetVT) {
if (OpVT == MVT::i32) {
if (RetVT == MVT::f32)
return UINTTOFP_I32_F32;
setOperationAction(ISD::PREFETCH, MVT::Other, Expand);
// ConstantFP nodes default to expand. Targets can either change this to
- // Legal, in which case all fp constants are legal, or use addLegalFPImmediate
+ // Legal, in which case all fp constants are legal, or use isFPImmLegal()
// to optimize expansions for certain constants.
setOperationAction(ISD::ConstantFP, MVT::f32, Expand);
setOperationAction(ISD::ConstantFP, MVT::f64, Expand);
setOperationAction(ISD::TRAP, MVT::Other, Expand);
IsLittleEndian = TD->isLittleEndian();
- UsesGlobalOffsetTable = false;
- ShiftAmountTy = PointerTy = getValueType(TD->getIntPtrType()).getSimpleVT();
+ ShiftAmountTy = PointerTy = MVT::getIntegerVT(8*TD->getPointerSize());
memset(RegClassForVT, 0,MVT::LAST_VALUETYPE*sizeof(TargetRegisterClass*));
memset(TargetDAGCombineArray, 0, array_lengthof(TargetDAGCombineArray));
maxStoresPerMemset = maxStoresPerMemcpy = maxStoresPerMemmove = 8;
- allowUnalignedMemoryAccesses = false;
benefitFromCodePlacementOpt = false;
UseUnderscoreSetJmp = false;
UseUnderscoreLongJmp = false;
InitLibcallNames(LibcallRoutineNames);
InitCmpLibcallCCs(CmpLibcallCCs);
-
- // Tell Legalize whether the assembler supports DEBUG_LOC.
- const TargetAsmInfo *TASM = TM.getTargetAsmInfo();
- if (!TASM || !TASM->hasDotLocAndDotFile())
- setOperationAction(ISD::DEBUG_LOC, MVT::Other, Expand);
+ InitLibcallCallingConvs(LibcallCallingConvs);
}
TargetLowering::~TargetLowering() {
delete &TLOF;
}
+/// canOpTrap - Returns true if the operation can trap for the value type.
+/// VT must be a legal type.
+bool TargetLowering::canOpTrap(unsigned Op, EVT VT) const {
+ assert(isTypeLegal(VT));
+ switch (Op) {
+ default:
+ return false;
+ case ISD::FDIV:
+ case ISD::FREM:
+ case ISD::SDIV:
+ case ISD::UDIV:
+ case ISD::SREM:
+ case ISD::UREM:
+ return true;
+ }
+}
+
+
+static unsigned getVectorTypeBreakdownMVT(MVT VT, MVT &IntermediateVT,
+ unsigned &NumIntermediates,
+ EVT &RegisterVT,
+ TargetLowering* TLI) {
+ // Figure out the right, legal destination reg to copy into.
+ unsigned NumElts = VT.getVectorNumElements();
+ MVT EltTy = VT.getVectorElementType();
+
+ unsigned NumVectorRegs = 1;
+
+ // FIXME: We don't support non-power-of-2-sized vectors for now. Ideally we
+ // could break down into LHS/RHS like LegalizeDAG does.
+ if (!isPowerOf2_32(NumElts)) {
+ NumVectorRegs = NumElts;
+ NumElts = 1;
+ }
+
+ // Divide the input until we get to a supported size. This will always
+ // end with a scalar if the target doesn't support vectors.
+ while (NumElts > 1 && !TLI->isTypeLegal(MVT::getVectorVT(EltTy, NumElts))) {
+ NumElts >>= 1;
+ NumVectorRegs <<= 1;
+ }
+
+ NumIntermediates = NumVectorRegs;
+
+ MVT NewVT = MVT::getVectorVT(EltTy, NumElts);
+ if (!TLI->isTypeLegal(NewVT))
+ NewVT = EltTy;
+ IntermediateVT = NewVT;
+
+ EVT DestVT = TLI->getRegisterType(NewVT);
+ RegisterVT = DestVT;
+ if (EVT(DestVT).bitsLT(NewVT)) {
+ // Value is expanded, e.g. i64 -> i16.
+ return NumVectorRegs*(NewVT.getSizeInBits()/DestVT.getSizeInBits());
+ } else {
+ // Otherwise, promotion or legal types use the same number of registers as
+ // the vector decimated to the appropriate level.
+ return NumVectorRegs;
+ }
+
+ return 1;
+}
+
/// computeRegisterProperties - Once all of the register classes are added,
/// this allows us to compute derived properties we expose.
void TargetLowering::computeRegisterProperties() {
// Every integer value type larger than this largest register takes twice as
// many registers to represent as the previous ValueType.
for (unsigned ExpandedReg = LargestIntReg + 1; ; ++ExpandedReg) {
- MVT EVT = (MVT::SimpleValueType)ExpandedReg;
- if (!EVT.isInteger())
+ EVT ExpandedVT = (MVT::SimpleValueType)ExpandedReg;
+ if (!ExpandedVT.isInteger())
break;
NumRegistersForVT[ExpandedReg] = 2*NumRegistersForVT[ExpandedReg-1];
RegisterTypeForVT[ExpandedReg] = (MVT::SimpleValueType)LargestIntReg;
TransformToType[ExpandedReg] = (MVT::SimpleValueType)(ExpandedReg - 1);
- ValueTypeActions.setTypeAction(EVT, Expand);
+ ValueTypeActions.setTypeAction(ExpandedVT, Expand);
}
// Inspect all of the ValueType's smaller than the largest integer
unsigned LegalIntReg = LargestIntReg;
for (unsigned IntReg = LargestIntReg - 1;
IntReg >= (unsigned)MVT::i1; --IntReg) {
- MVT IVT = (MVT::SimpleValueType)IntReg;
+ EVT IVT = (MVT::SimpleValueType)IntReg;
if (isTypeLegal(IVT)) {
LegalIntReg = IntReg;
} else {
i <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++i) {
MVT VT = (MVT::SimpleValueType)i;
if (!isTypeLegal(VT)) {
- MVT IntermediateVT, RegisterVT;
+ MVT IntermediateVT;
+ EVT RegisterVT;
unsigned NumIntermediates;
NumRegistersForVT[i] =
- getVectorTypeBreakdown(VT,
- IntermediateVT, NumIntermediates,
- RegisterVT);
+ getVectorTypeBreakdownMVT(VT, IntermediateVT, NumIntermediates,
+ RegisterVT, this);
RegisterTypeForVT[i] = RegisterVT;
// Determine if there is a legal wider type.
bool IsLegalWiderType = false;
- MVT EltVT = VT.getVectorElementType();
+ EVT EltVT = VT.getVectorElementType();
unsigned NElts = VT.getVectorNumElements();
for (unsigned nVT = i+1; nVT <= MVT::LAST_VECTOR_VALUETYPE; ++nVT) {
- MVT SVT = (MVT::SimpleValueType)nVT;
+ EVT SVT = (MVT::SimpleValueType)nVT;
if (isTypeLegal(SVT) && SVT.getVectorElementType() == EltVT &&
- SVT.getVectorNumElements() > NElts) {
+ SVT.getVectorNumElements() > NElts && NElts != 1) {
TransformToType[i] = SVT;
ValueTypeActions.setTypeAction(VT, Promote);
IsLegalWiderType = true;
}
}
if (!IsLegalWiderType) {
- MVT NVT = VT.getPow2VectorType();
+ EVT NVT = VT.getPow2VectorType();
if (NVT == VT) {
// Type is already a power of 2. The default action is to split.
TransformToType[i] = MVT::Other;
}
-MVT::SimpleValueType TargetLowering::getSetCCResultType(MVT VT) const {
- return getValueType(TD->getIntPtrType()).getSimpleVT();
+MVT::SimpleValueType TargetLowering::getSetCCResultType(EVT VT) const {
+ return PointerTy.SimpleTy;
}
+MVT::SimpleValueType TargetLowering::getCmpLibcallReturnType() const {
+ return MVT::i32; // return the default value
+}
/// getVectorTypeBreakdown - Vector types are broken down into some number of
/// legal first class types. For example, MVT::v8f32 maps to 2 MVT::v4f32
/// register. It also returns the VT and quantity of the intermediate values
/// before they are promoted/expanded.
///
-unsigned TargetLowering::getVectorTypeBreakdown(MVT VT,
- MVT &IntermediateVT,
+unsigned TargetLowering::getVectorTypeBreakdown(LLVMContext &Context, EVT VT,
+ EVT &IntermediateVT,
unsigned &NumIntermediates,
- MVT &RegisterVT) const {
+ EVT &RegisterVT) const {
// Figure out the right, legal destination reg to copy into.
unsigned NumElts = VT.getVectorNumElements();
- MVT EltTy = VT.getVectorElementType();
+ EVT EltTy = VT.getVectorElementType();
unsigned NumVectorRegs = 1;
// Divide the input until we get to a supported size. This will always
// end with a scalar if the target doesn't support vectors.
- while (NumElts > 1 && !isTypeLegal(MVT::getVectorVT(EltTy, NumElts))) {
+ while (NumElts > 1 && !isTypeLegal(
+ EVT::getVectorVT(Context, EltTy, NumElts))) {
NumElts >>= 1;
NumVectorRegs <<= 1;
}
NumIntermediates = NumVectorRegs;
- MVT NewVT = MVT::getVectorVT(EltTy, NumElts);
+ EVT NewVT = EVT::getVectorVT(Context, EltTy, NumElts);
if (!isTypeLegal(NewVT))
NewVT = EltTy;
IntermediateVT = NewVT;
- MVT DestVT = getRegisterType(NewVT);
+ EVT DestVT = getRegisterType(Context, NewVT);
RegisterVT = DestVT;
if (DestVT.bitsLT(NewVT)) {
// Value is expanded, e.g. i64 -> i16.
return 1;
}
-/// getWidenVectorType: given a vector type, returns the type to widen to
-/// (e.g., v7i8 to v8i8). If the vector type is legal, it returns itself.
-/// If there is no vector type that we want to widen to, returns MVT::Other
-/// When and where to widen is target dependent based on the cost of
-/// scalarizing vs using the wider vector type.
-MVT TargetLowering::getWidenVectorType(MVT VT) const {
- assert(VT.isVector());
- if (isTypeLegal(VT))
- return VT;
-
- // Default is not to widen until moved to LegalizeTypes
- return MVT::Other;
-}
-
/// getByValTypeAlignment - Return the desired alignment for ByVal aggregate
/// function arguments in the caller parameter area. This is the actual
/// alignment, not its logarithm.
return TD->getCallFrameTypeAlignment(Ty);
}
+/// getJumpTableEncoding - Return the entry encoding for a jump table in the
+/// current function. The returned value is a member of the
+/// MachineJumpTableInfo::JTEntryKind enum.
+unsigned TargetLowering::getJumpTableEncoding() const {
+ // In non-pic modes, just use the address of a block.
+ if (getTargetMachine().getRelocationModel() != Reloc::PIC_)
+ return MachineJumpTableInfo::EK_BlockAddress;
+
+ // In PIC mode, if the target supports a GPRel32 directive, use it.
+ if (getTargetMachine().getMCAsmInfo()->getGPRel32Directive() != 0)
+ return MachineJumpTableInfo::EK_GPRel32BlockAddress;
+
+ // Otherwise, use a label difference.
+ return MachineJumpTableInfo::EK_LabelDifference32;
+}
+
SDValue TargetLowering::getPICJumpTableRelocBase(SDValue Table,
SelectionDAG &DAG) const {
- if (usesGlobalOffsetTable())
+ // If our PIC model is GP relative, use the global offset table as the base.
+ if (getJumpTableEncoding() == MachineJumpTableInfo::EK_GPRel32BlockAddress)
return DAG.getGLOBAL_OFFSET_TABLE(getPointerTy());
return Table;
}
+/// getPICJumpTableRelocBaseExpr - This returns the relocation base for the
+/// given PIC jumptable, the same as getPICJumpTableRelocBase, but as an
+/// MCExpr.
+const MCExpr *
+TargetLowering::getPICJumpTableRelocBaseExpr(const MachineFunction *MF,
+ unsigned JTI,MCContext &Ctx) const{
+ // The normal PIC reloc base is the label at the start of the jump table.
+ return MCSymbolRefExpr::Create(MF->getJTISymbol(JTI, Ctx), Ctx);
+}
+
bool
TargetLowering::isOffsetFoldingLegal(const GlobalAddressSDNode *GA) const {
// Assume that everything is safe in static mode.
// if we can expand it to have all bits set, do it
if (C->getAPIntValue().intersects(~Demanded)) {
- MVT VT = Op.getValueType();
+ EVT VT = Op.getValueType();
SDValue New = DAG.getNode(Op.getOpcode(), dl, VT, Op.getOperand(0),
DAG.getConstant(Demanded &
C->getAPIntValue(),
if (!isPowerOf2_32(SmallVTBits))
SmallVTBits = NextPowerOf2(SmallVTBits);
for (; SmallVTBits < BitWidth; SmallVTBits = NextPowerOf2(SmallVTBits)) {
- MVT SmallVT = MVT::getIntegerVT(SmallVTBits);
+ EVT SmallVT = EVT::getIntegerVT(*DAG.getContext(), SmallVTBits);
if (TLI.isTruncateFree(Op.getValueType(), SmallVT) &&
TLI.isZExtFree(SmallVT, Op.getValueType())) {
// We found a type with free casts.
TargetLoweringOpt &TLO,
unsigned Depth) const {
unsigned BitWidth = DemandedMask.getBitWidth();
- assert(Op.getValueSizeInBits() == BitWidth &&
+ assert(Op.getValueType().getScalarType().getSizeInBits() == BitWidth &&
"Mask size mismatches value type size!");
APInt NewMask = DemandedMask;
DebugLoc dl = Op.getDebugLoc();
if (TLO.ShrinkDemandedConstant(Op, ~KnownZero2 & NewMask))
return true;
// If the operation can be done in a smaller type, do so.
- if (TLO.ShrinkDemandedOp(Op, BitWidth, NewMask, dl))
+ if (TLO.ShrinkOps && TLO.ShrinkDemandedOp(Op, BitWidth, NewMask, dl))
return true;
// Output known-1 bits are only known if set in both the LHS & RHS.
if (TLO.ShrinkDemandedConstant(Op, NewMask))
return true;
// If the operation can be done in a smaller type, do so.
- if (TLO.ShrinkDemandedOp(Op, BitWidth, NewMask, dl))
+ if (TLO.ShrinkOps && TLO.ShrinkDemandedOp(Op, BitWidth, NewMask, dl))
return true;
// Output known-0 bits are only known if clear in both the LHS & RHS.
if ((KnownZero2 & NewMask) == NewMask)
return TLO.CombineTo(Op, Op.getOperand(1));
// If the operation can be done in a smaller type, do so.
- if (TLO.ShrinkDemandedOp(Op, BitWidth, NewMask, dl))
+ if (TLO.ShrinkOps && TLO.ShrinkDemandedOp(Op, BitWidth, NewMask, dl))
return true;
// If all of the unknown bits are known to be zero on one side or the other
// e.g. (X | C1) ^ C2 --> (X | C1) & ~C2 iff (C1&C2) == C2
if ((NewMask & (KnownZero|KnownOne)) == NewMask) { // all known
if ((KnownOne & KnownOne2) == KnownOne) {
- MVT VT = Op.getValueType();
+ EVT VT = Op.getValueType();
SDValue ANDC = TLO.DAG.getConstant(~KnownOne & NewMask, VT);
return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::AND, dl, VT,
Op.getOperand(0), ANDC));
// if we can expand it to have all bits set, do it
if (Expanded.isAllOnesValue()) {
if (Expanded != C->getAPIntValue()) {
- MVT VT = Op.getValueType();
+ EVT VT = Op.getValueType();
SDValue New = TLO.DAG.getNode(Op.getOpcode(), dl,VT, Op.getOperand(0),
TLO.DAG.getConstant(Expanded, VT));
return TLO.CombineTo(Op, New);
SDValue NewSA =
TLO.DAG.getConstant(Diff, Op.getOperand(1).getValueType());
- MVT VT = Op.getValueType();
+ EVT VT = Op.getValueType();
return TLO.CombineTo(Op, TLO.DAG.getNode(Opc, dl, VT,
InOp.getOperand(0), NewSA));
}
break;
case ISD::SRL:
if (ConstantSDNode *SA = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
- MVT VT = Op.getValueType();
+ EVT VT = Op.getValueType();
unsigned ShAmt = SA->getZExtValue();
unsigned VTSize = VT.getSizeInBits();
SDValue InOp = Op.getOperand(0);
Op.getOperand(0), Op.getOperand(1)));
if (ConstantSDNode *SA = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
- MVT VT = Op.getValueType();
+ EVT VT = Op.getValueType();
unsigned ShAmt = SA->getZExtValue();
// If the shift count is an invalid immediate, don't do anything.
// demand the input sign bit.
APInt HighBits = APInt::getHighBitsSet(BitWidth, ShAmt);
if (HighBits.intersects(NewMask))
- InDemandedMask |= APInt::getSignBit(VT.getSizeInBits());
+ InDemandedMask |= APInt::getSignBit(VT.getScalarType().getSizeInBits());
if (SimplifyDemandedBits(Op.getOperand(0), InDemandedMask,
KnownZero, KnownOne, TLO, Depth+1))
}
break;
case ISD::SIGN_EXTEND_INREG: {
- MVT EVT = cast<VTSDNode>(Op.getOperand(1))->getVT();
+ EVT EVT = cast<VTSDNode>(Op.getOperand(1))->getVT();
// Sign extension. Compute the demanded bits in the result that are not
// present in the input.
- APInt NewBits = APInt::getHighBitsSet(BitWidth,
- BitWidth - EVT.getSizeInBits()) &
- NewMask;
+ APInt NewBits =
+ APInt::getHighBitsSet(BitWidth,
+ BitWidth - EVT.getScalarType().getSizeInBits()) &
+ NewMask;
// If none of the extended bits are demanded, eliminate the sextinreg.
if (NewBits == 0)
return TLO.CombineTo(Op, Op.getOperand(0));
- APInt InSignBit = APInt::getSignBit(EVT.getSizeInBits());
+ APInt InSignBit = APInt::getSignBit(EVT.getScalarType().getSizeInBits());
InSignBit.zext(BitWidth);
- APInt InputDemandedBits = APInt::getLowBitsSet(BitWidth,
- EVT.getSizeInBits()) &
- NewMask;
+ APInt InputDemandedBits =
+ APInt::getLowBitsSet(BitWidth,
+ EVT.getScalarType().getSizeInBits()) &
+ NewMask;
// Since the sign extended bits are demanded, we know that the sign
// bit is demanded.
break;
}
case ISD::ZERO_EXTEND: {
- unsigned OperandBitWidth = Op.getOperand(0).getValueSizeInBits();
+ unsigned OperandBitWidth =
+ Op.getOperand(0).getValueType().getScalarType().getSizeInBits();
APInt InMask = NewMask;
InMask.trunc(OperandBitWidth);
break;
}
case ISD::SIGN_EXTEND: {
- MVT InVT = Op.getOperand(0).getValueType();
- unsigned InBits = InVT.getSizeInBits();
+ EVT InVT = Op.getOperand(0).getValueType();
+ unsigned InBits = InVT.getScalarType().getSizeInBits();
APInt InMask = APInt::getLowBitsSet(BitWidth, InBits);
APInt InSignBit = APInt::getBitsSet(BitWidth, InBits - 1, InBits);
APInt NewBits = ~InMask & NewMask;
break;
}
case ISD::ANY_EXTEND: {
- unsigned OperandBitWidth = Op.getOperand(0).getValueSizeInBits();
+ unsigned OperandBitWidth =
+ Op.getOperand(0).getValueType().getScalarType().getSizeInBits();
APInt InMask = NewMask;
InMask.trunc(OperandBitWidth);
if (SimplifyDemandedBits(Op.getOperand(0), InMask,
case ISD::TRUNCATE: {
// Simplify the input, using demanded bit information, and compute the known
// zero/one bits live out.
+ unsigned OperandBitWidth =
+ Op.getOperand(0).getValueType().getScalarType().getSizeInBits();
APInt TruncMask = NewMask;
- TruncMask.zext(Op.getOperand(0).getValueSizeInBits());
+ TruncMask.zext(OperandBitWidth);
if (SimplifyDemandedBits(Op.getOperand(0), TruncMask,
KnownZero, KnownOne, TLO, Depth+1))
return true;
// on the known demanded bits.
if (Op.getOperand(0).getNode()->hasOneUse()) {
SDValue In = Op.getOperand(0);
- unsigned InBitWidth = In.getValueSizeInBits();
switch (In.getOpcode()) {
default: break;
case ISD::SRL:
// Shrink SRL by a constant if none of the high bits shifted in are
// demanded.
if (ConstantSDNode *ShAmt = dyn_cast<ConstantSDNode>(In.getOperand(1))){
- APInt HighBits = APInt::getHighBitsSet(InBitWidth,
- InBitWidth - BitWidth);
+ APInt HighBits = APInt::getHighBitsSet(OperandBitWidth,
+ OperandBitWidth - BitWidth);
HighBits = HighBits.lshr(ShAmt->getZExtValue());
HighBits.trunc(BitWidth);
break;
}
case ISD::AssertZext: {
- MVT VT = cast<VTSDNode>(Op.getOperand(1))->getVT();
+ EVT VT = cast<VTSDNode>(Op.getOperand(1))->getVT();
APInt InMask = APInt::getLowBitsSet(BitWidth,
VT.getSizeInBits());
if (SimplifyDemandedBits(Op.getOperand(0), InMask & NewMask,
#if 0
// If this is an FP->Int bitcast and if the sign bit is the only thing that
// is demanded, turn this into a FGETSIGN.
- if (NewMask == MVT::getIntegerVTSignBit(Op.getValueType()) &&
+ if (NewMask == EVT::getIntegerVTSignBit(Op.getValueType()) &&
MVT::isFloatingPoint(Op.getOperand(0).getValueType()) &&
!MVT::isVector(Op.getOperand(0).getValueType())) {
// Only do this xform if FGETSIGN is valid or if before legalize.
KnownOne2, TLO, Depth+1))
return true;
// See if the operation should be performed at a smaller bit width.
- if (TLO.ShrinkDemandedOp(Op, BitWidth, NewMask, dl))
+ if (TLO.ShrinkOps && TLO.ShrinkDemandedOp(Op, BitWidth, NewMask, dl))
return true;
}
// FALL THROUGH
// to handle some common cases.
// Fall back to ComputeMaskedBits to catch other known cases.
- MVT OpVT = Val.getValueType();
- unsigned BitWidth = OpVT.getSizeInBits();
+ EVT OpVT = Val.getValueType();
+ unsigned BitWidth = OpVT.getScalarType().getSizeInBits();
APInt Mask = APInt::getAllOnesValue(BitWidth);
APInt KnownZero, KnownOne;
DAG.ComputeMaskedBits(Val, Mask, KnownZero, KnownOne);
/// SimplifySetCC - Try to simplify a setcc built with the specified operands
/// and cc. If it is unable to simplify it, return a null SDValue.
SDValue
-TargetLowering::SimplifySetCC(MVT VT, SDValue N0, SDValue N1,
+TargetLowering::SimplifySetCC(EVT VT, SDValue N0, SDValue N1,
ISD::CondCode Cond, bool foldBooleans,
DAGCombinerInfo &DCI, DebugLoc dl) const {
SelectionDAG &DAG = DCI.DAG;
+ LLVMContext &Context = *DAG.getContext();
// These setcc operations always fold.
switch (Cond) {
if (N0.getOpcode() == ISD::SRL && (C1 == 0 || C1 == 1) &&
N0.getOperand(0).getOpcode() == ISD::CTLZ &&
N0.getOperand(1).getOpcode() == ISD::Constant) {
- unsigned ShAmt = cast<ConstantSDNode>(N0.getOperand(1))->getZExtValue();
+ const APInt &ShAmt
+ = cast<ConstantSDNode>(N0.getOperand(1))->getAPIntValue();
if ((Cond == ISD::SETEQ || Cond == ISD::SETNE) &&
ShAmt == Log2_32(N0.getValueType().getSizeInBits())) {
if ((C1 == 0) == (Cond == ISD::SETEQ)) {
N0.getOperand(0).getNode()->hasOneUse() &&
isa<ConstantSDNode>(N0.getOperand(1))) {
LoadSDNode *Lod = cast<LoadSDNode>(N0.getOperand(0));
- uint64_t bestMask = 0;
+ APInt bestMask;
unsigned bestWidth = 0, bestOffset = 0;
- if (!Lod->isVolatile() && Lod->isUnindexed() &&
- // FIXME: This uses getZExtValue() below so it only works on i64 and
- // below.
- N0.getValueType().getSizeInBits() <= 64) {
+ if (!Lod->isVolatile() && Lod->isUnindexed()) {
unsigned origWidth = N0.getValueType().getSizeInBits();
+ unsigned maskWidth = origWidth;
// We can narrow (e.g.) 16-bit extending loads on 32-bit target to
// 8 bits, but have to be careful...
if (Lod->getExtensionType() != ISD::NON_EXTLOAD)
origWidth = Lod->getMemoryVT().getSizeInBits();
- uint64_t Mask =cast<ConstantSDNode>(N0.getOperand(1))->getZExtValue();
+ const APInt &Mask =
+ cast<ConstantSDNode>(N0.getOperand(1))->getAPIntValue();
for (unsigned width = origWidth / 2; width>=8; width /= 2) {
- uint64_t newMask = (1ULL << width) - 1;
+ APInt newMask = APInt::getLowBitsSet(maskWidth, width);
for (unsigned offset=0; offset<origWidth/width; offset++) {
if ((newMask & Mask) == Mask) {
if (!TD->isLittleEndian())
bestOffset = (origWidth/width - offset - 1) * (width/8);
else
bestOffset = (uint64_t)offset * (width/8);
- bestMask = Mask >> (offset * (width/8) * 8);
+ bestMask = Mask.lshr(offset * (width/8) * 8);
bestWidth = width;
break;
}
}
}
if (bestWidth) {
- MVT newVT = MVT::getIntegerVT(bestWidth);
+ EVT newVT = EVT::getIntegerVT(Context, bestWidth);
if (newVT.isRound()) {
- MVT PtrType = Lod->getOperand(1).getValueType();
+ EVT PtrType = Lod->getOperand(1).getValueType();
SDValue Ptr = Lod->getBasePtr();
if (bestOffset != 0)
Ptr = DAG.getNode(ISD::ADD, dl, PtrType, Lod->getBasePtr(),
SDValue NewLoad = DAG.getLoad(newVT, dl, Lod->getChain(), Ptr,
Lod->getSrcValue(),
Lod->getSrcValueOffset() + bestOffset,
- false, NewAlign);
+ false, false, NewAlign);
return DAG.getSetCC(dl, VT,
DAG.getNode(ISD::AND, dl, newVT, NewLoad,
- DAG.getConstant(bestMask, newVT)),
+ DAG.getConstant(bestMask.trunc(bestWidth),
+ newVT)),
DAG.getConstant(0LL, newVT), Cond);
}
}
case ISD::SETUGE:
case ISD::SETULT:
case ISD::SETULE: {
- MVT newVT = N0.getOperand(0).getValueType();
+ EVT newVT = N0.getOperand(0).getValueType();
if (DCI.isBeforeLegalizeOps() ||
(isOperationLegal(ISD::SETCC, newVT) &&
getCondCodeAction(Cond, newVT)==Legal))
break; // todo, be more careful with signed comparisons
}
} else if (N0.getOpcode() == ISD::SIGN_EXTEND_INREG &&
- (Cond == ISD::SETEQ || Cond == ISD::SETNE)) {
- MVT ExtSrcTy = cast<VTSDNode>(N0.getOperand(1))->getVT();
+ (Cond == ISD::SETEQ || Cond == ISD::SETNE)) {
+ EVT ExtSrcTy = cast<VTSDNode>(N0.getOperand(1))->getVT();
unsigned ExtSrcTyBits = ExtSrcTy.getSizeInBits();
- MVT ExtDstTy = N0.getValueType();
+ EVT ExtDstTy = N0.getValueType();
unsigned ExtDstTyBits = ExtDstTy.getSizeInBits();
// If the extended part has any inconsistent bits, it cannot ever
return DAG.getConstant(Cond == ISD::SETNE, VT);
SDValue ZextOp;
- MVT Op0Ty = N0.getOperand(0).getValueType();
+ EVT Op0Ty = N0.getOperand(0).getValueType();
if (Op0Ty == ExtSrcTy) {
ZextOp = N0.getOperand(0);
} else {
Cond);
} else if ((N1C->isNullValue() || N1C->getAPIntValue() == 1) &&
(Cond == ISD::SETEQ || Cond == ISD::SETNE)) {
-
// SETCC (SETCC), [0|1], [EQ|NE] -> SETCC
- if (N0.getOpcode() == ISD::SETCC) {
- bool TrueWhenTrue = (Cond == ISD::SETEQ) ^ (N1C->getZExtValue() != 1);
+ if (N0.getOpcode() == ISD::SETCC &&
+ isTypeLegal(VT) && VT.bitsLE(N0.getValueType())) {
+ bool TrueWhenTrue = (Cond == ISD::SETEQ) ^ (N1C->getAPIntValue() != 1);
if (TrueWhenTrue)
- return N0;
-
+ return DAG.getNode(ISD::TRUNCATE, dl, VT, N0);
// Invert the condition.
ISD::CondCode CC = cast<CondCodeSDNode>(N0.getOperand(2))->get();
CC = ISD::getSetCCInverse(CC,
N0.getOperand(0).getValueType().isInteger());
return DAG.getSetCC(dl, VT, N0.getOperand(0), N0.getOperand(1), CC);
}
-
+
if ((N0.getOpcode() == ISD::XOR ||
- (N0.getOpcode() == ISD::AND &&
+ (N0.getOpcode() == ISD::AND &&
N0.getOperand(0).getOpcode() == ISD::XOR &&
N0.getOperand(1) == N0.getOperand(0).getOperand(1))) &&
isa<ConstantSDNode>(N0.getOperand(1)) &&
N0.getOperand(0).getOperand(0),
N0.getOperand(1));
}
+
return DAG.getSetCC(dl, VT, Val, N1,
Cond == ISD::SETEQ ? ISD::SETNE : ISD::SETEQ);
}
+ } else if (N1C->getAPIntValue() == 1 &&
+ (VT == MVT::i1 ||
+ getBooleanContents() == ZeroOrOneBooleanContent)) {
+ SDValue Op0 = N0;
+ if (Op0.getOpcode() == ISD::TRUNCATE)
+ Op0 = Op0.getOperand(0);
+
+ if ((Op0.getOpcode() == ISD::XOR) &&
+ Op0.getOperand(0).getOpcode() == ISD::SETCC &&
+ Op0.getOperand(1).getOpcode() == ISD::SETCC) {
+ // (xor (setcc), (setcc)) == / != 1 -> (setcc) != / == (setcc)
+ Cond = (Cond == ISD::SETEQ) ? ISD::SETNE : ISD::SETEQ;
+ return DAG.getSetCC(dl, VT, Op0.getOperand(0), Op0.getOperand(1),
+ Cond);
+ } else if (Op0.getOpcode() == ISD::AND &&
+ isa<ConstantSDNode>(Op0.getOperand(1)) &&
+ cast<ConstantSDNode>(Op0.getOperand(1))->getAPIntValue() == 1) {
+ // If this is (X&1) == / != 1, normalize it to (X&1) != / == 0.
+ if (Op0.getValueType() != VT)
+ Op0 = DAG.getNode(ISD::AND, dl, VT,
+ DAG.getNode(ISD::TRUNCATE, dl, VT, Op0.getOperand(0)),
+ DAG.getConstant(1, VT));
+ return DAG.getSetCC(dl, VT, Op0,
+ DAG.getConstant(0, Op0.getValueType()),
+ Cond == ISD::SETEQ ? ISD::SETNE : ISD::SETEQ);
+ }
}
}
// Fold bit comparisons when we can.
if ((Cond == ISD::SETEQ || Cond == ISD::SETNE) &&
- VT == N0.getValueType() && N0.getOpcode() == ISD::AND)
+ (VT == N0.getValueType() ||
+ (isTypeLegal(VT) && VT.bitsLE(N0.getValueType()))) &&
+ N0.getOpcode() == ISD::AND)
if (ConstantSDNode *AndRHS =
dyn_cast<ConstantSDNode>(N0.getOperand(1))) {
- MVT ShiftTy = DCI.isBeforeLegalize() ?
+ EVT ShiftTy = DCI.isBeforeLegalize() ?
getPointerTy() : getShiftAmountTy();
if (Cond == ISD::SETNE && C1 == 0) {// (X & 8) != 0 --> (X & 8) >> 3
// Perform the xform if the AND RHS is a single bit.
- if (isPowerOf2_64(AndRHS->getZExtValue())) {
- return DAG.getNode(ISD::SRL, dl, VT, N0,
- DAG.getConstant(Log2_64(AndRHS->getZExtValue()),
- ShiftTy));
+ if (AndRHS->getAPIntValue().isPowerOf2()) {
+ return DAG.getNode(ISD::TRUNCATE, dl, VT,
+ DAG.getNode(ISD::SRL, dl, N0.getValueType(), N0,
+ DAG.getConstant(AndRHS->getAPIntValue().logBase2(), ShiftTy)));
}
- } else if (Cond == ISD::SETEQ && C1 == AndRHS->getZExtValue()) {
+ } else if (Cond == ISD::SETEQ && C1 == AndRHS->getAPIntValue()) {
// (X & 8) == 8 --> (X & 8) >> 3
// Perform the xform if C1 is a single bit.
if (C1.isPowerOf2()) {
- return DAG.getNode(ISD::SRL, dl, VT, N0,
- DAG.getConstant(C1.logBase2(), ShiftTy));
+ return DAG.getNode(ISD::TRUNCATE, dl, VT,
+ DAG.getNode(ISD::SRL, dl, N0.getValueType(), N0,
+ DAG.getConstant(C1.logBase2(), ShiftTy)));
}
}
}
// materialize 0.0.
if (Cond == ISD::SETO || Cond == ISD::SETUO)
return DAG.getSetCC(dl, VT, N0, N0, Cond);
+
+ // If the condition is not legal, see if we can find an equivalent one
+ // which is legal.
+ if (!isCondCodeLegal(Cond, N0.getValueType())) {
+ // If the comparison was an awkward floating-point == or != and one of
+ // the comparison operands is infinity or negative infinity, convert the
+ // condition to a less-awkward <= or >=.
+ if (CFP->getValueAPF().isInfinity()) {
+ if (CFP->getValueAPF().isNegative()) {
+ if (Cond == ISD::SETOEQ &&
+ isCondCodeLegal(ISD::SETOLE, N0.getValueType()))
+ return DAG.getSetCC(dl, VT, N0, N1, ISD::SETOLE);
+ if (Cond == ISD::SETUEQ &&
+ isCondCodeLegal(ISD::SETOLE, N0.getValueType()))
+ return DAG.getSetCC(dl, VT, N0, N1, ISD::SETULE);
+ if (Cond == ISD::SETUNE &&
+ isCondCodeLegal(ISD::SETUGT, N0.getValueType()))
+ return DAG.getSetCC(dl, VT, N0, N1, ISD::SETUGT);
+ if (Cond == ISD::SETONE &&
+ isCondCodeLegal(ISD::SETUGT, N0.getValueType()))
+ return DAG.getSetCC(dl, VT, N0, N1, ISD::SETOGT);
+ } else {
+ if (Cond == ISD::SETOEQ &&
+ isCondCodeLegal(ISD::SETOGE, N0.getValueType()))
+ return DAG.getSetCC(dl, VT, N0, N1, ISD::SETOGE);
+ if (Cond == ISD::SETUEQ &&
+ isCondCodeLegal(ISD::SETOGE, N0.getValueType()))
+ return DAG.getSetCC(dl, VT, N0, N1, ISD::SETUGE);
+ if (Cond == ISD::SETUNE &&
+ isCondCodeLegal(ISD::SETULT, N0.getValueType()))
+ return DAG.getSetCC(dl, VT, N0, N1, ISD::SETULT);
+ if (Cond == ISD::SETONE &&
+ isCondCodeLegal(ISD::SETULT, N0.getValueType()))
+ return DAG.getSetCC(dl, VT, N0, N1, ISD::SETOLT);
+ }
+ }
+ }
}
if (N0 == N1) {
}
-/// isConsecutiveLoad - Return true if LD is loading 'Bytes' bytes from a
-/// location that is 'Dist' units away from the location that the 'Base' load
-/// is loading from.
-bool TargetLowering::isConsecutiveLoad(LoadSDNode *LD, LoadSDNode *Base,
- unsigned Bytes, int Dist,
- const MachineFrameInfo *MFI) const {
- if (LD->getChain() != Base->getChain())
- return false;
- MVT VT = LD->getValueType(0);
- if (VT.getSizeInBits() / 8 != Bytes)
- return false;
-
- SDValue Loc = LD->getOperand(1);
- SDValue BaseLoc = Base->getOperand(1);
- if (Loc.getOpcode() == ISD::FrameIndex) {
- if (BaseLoc.getOpcode() != ISD::FrameIndex)
- return false;
- int FI = cast<FrameIndexSDNode>(Loc)->getIndex();
- int BFI = cast<FrameIndexSDNode>(BaseLoc)->getIndex();
- int FS = MFI->getObjectSize(FI);
- int BFS = MFI->getObjectSize(BFI);
- if (FS != BFS || FS != (int)Bytes) return false;
- return MFI->getObjectOffset(FI) == (MFI->getObjectOffset(BFI) + Dist*Bytes);
- }
- if (Loc.getOpcode() == ISD::ADD && Loc.getOperand(0) == BaseLoc) {
- ConstantSDNode *V = dyn_cast<ConstantSDNode>(Loc.getOperand(1));
- if (V && (V->getSExtValue() == Dist*Bytes))
- return true;
- }
-
- GlobalValue *GV1 = NULL;
- GlobalValue *GV2 = NULL;
- int64_t Offset1 = 0;
- int64_t Offset2 = 0;
- bool isGA1 = isGAPlusOffset(Loc.getNode(), GV1, Offset1);
- bool isGA2 = isGAPlusOffset(BaseLoc.getNode(), GV2, Offset2);
- if (isGA1 && isGA2 && GV1 == GV2)
- return Offset1 == (Offset2 + Dist*Bytes);
- return false;
-}
-
-
SDValue TargetLowering::
PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const {
// Default implementation: no optimization.
/// LowerXConstraint - try to replace an X constraint, which matches anything,
/// with another that has more specific requirements based on the type of the
/// corresponding operand.
-const char *TargetLowering::LowerXConstraint(MVT ConstraintVT) const{
+const char *TargetLowering::LowerXConstraint(EVT ConstraintVT) const{
if (ConstraintVT.isInteger())
return "r";
if (ConstraintVT.isFloatingPoint())
std::vector<unsigned> TargetLowering::
getRegClassForInlineAsmConstraint(const std::string &Constraint,
- MVT VT) const {
+ EVT VT) const {
return std::vector<unsigned>();
}
std::pair<unsigned, const TargetRegisterClass*> TargetLowering::
getRegForInlineAsmConstraint(const std::string &Constraint,
- MVT VT) const {
+ EVT VT) const {
if (Constraint[0] != '{')
return std::pair<unsigned, const TargetRegisterClass*>(0, 0);
assert(*(Constraint.end()-1) == '}' && "Not a brace enclosed constraint?");
// Remove the braces from around the name.
- std::string RegName(Constraint.begin()+1, Constraint.end()-1);
+ StringRef RegName(Constraint.data()+1, Constraint.size()-2);
// Figure out which register class contains this reg.
const TargetRegisterInfo *RI = TM.getRegisterInfo();
E = RI->regclass_end(); RCI != E; ++RCI) {
const TargetRegisterClass *RC = *RCI;
- // If none of the the value types for this register class are valid, we
+ // If none of the value types for this register class are valid, we
// can't use it. For example, 64-bit reg classes on 32-bit targets.
bool isLegal = false;
for (TargetRegisterClass::vt_iterator I = RC->vt_begin(), E = RC->vt_end();
for (TargetRegisterClass::iterator I = RC->begin(), E = RC->end();
I != E; ++I) {
- if (StringsEqualNoCase(RegName, RI->get(*I).AsmName))
+ if (RegName.equals_lower(RI->getName(*I)))
return std::make_pair(*I, RC);
}
}
/// <http://the.wall.riscom.net/books/proc/ppc/cwg/code2.html>
SDValue TargetLowering::BuildSDIV(SDNode *N, SelectionDAG &DAG,
std::vector<SDNode*>* Created) const {
- MVT VT = N->getValueType(0);
+ EVT VT = N->getValueType(0);
DebugLoc dl= N->getDebugLoc();
// Check to see if we can do this.
/// <http://the.wall.riscom.net/books/proc/ppc/cwg/code2.html>
SDValue TargetLowering::BuildUDIV(SDNode *N, SelectionDAG &DAG,
std::vector<SDNode*>* Created) const {
- MVT VT = N->getValueType(0);
+ EVT VT = N->getValueType(0);
DebugLoc dl = N->getDebugLoc();
// Check to see if we can do this.
projects / oota-llvm.git / blobdiff