/// - the promotion of vector elements. This feature is disabled by default
/// and only enabled using this flag.
static cl::opt<bool>
-AllowPromoteIntElem("promote-elements", cl::Hidden,
+AllowPromoteIntElem("promote-elements", cl::Hidden, cl::init(true),
cl::desc("Allow promotion of integer vector element types"));
namespace llvm {
Names[RTLIB::MUL_I32] = "__mulsi3";
Names[RTLIB::MUL_I64] = "__muldi3";
Names[RTLIB::MUL_I128] = "__multi3";
+ Names[RTLIB::MULO_I32] = "__mulosi4";
+ Names[RTLIB::MULO_I64] = "__mulodi4";
+ Names[RTLIB::MULO_I128] = "__muloti4";
Names[RTLIB::SDIV_I8] = "__divqi3";
Names[RTLIB::SDIV_I16] = "__divhi3";
Names[RTLIB::SDIV_I32] = "__divsi3";
Names[RTLIB::REM_F64] = "fmod";
Names[RTLIB::REM_F80] = "fmodl";
Names[RTLIB::REM_PPCF128] = "fmodl";
+ Names[RTLIB::FMA_F32] = "fmaf";
+ Names[RTLIB::FMA_F64] = "fma";
+ Names[RTLIB::FMA_F80] = "fmal";
+ Names[RTLIB::FMA_PPCF128] = "fmal";
Names[RTLIB::POWI_F32] = "__powisf2";
Names[RTLIB::POWI_F64] = "__powidf2";
Names[RTLIB::POWI_F80] = "__powixf2";
Names[RTLIB::SYNC_FETCH_AND_OR_8] = "__sync_fetch_and_or_8";
Names[RTLIB::SYNC_FETCH_AND_XOR_1] = "__sync_fetch_and_xor_1";
Names[RTLIB::SYNC_FETCH_AND_XOR_2] = "__sync_fetch_and_xor_2";
- Names[RTLIB::SYNC_FETCH_AND_XOR_4] = "__sync_fetch_and-xor_4";
+ Names[RTLIB::SYNC_FETCH_AND_XOR_4] = "__sync_fetch_and_xor_4";
Names[RTLIB::SYNC_FETCH_AND_XOR_8] = "__sync_fetch_and_xor_8";
Names[RTLIB::SYNC_FETCH_AND_NAND_1] = "__sync_fetch_and_nand_1";
Names[RTLIB::SYNC_FETCH_AND_NAND_2] = "__sync_fetch_and_nand_2";
ExceptionPointerRegister = 0;
ExceptionSelectorRegister = 0;
BooleanContents = UndefinedBooleanContent;
- SchedPreferenceInfo = Sched::Latency;
+ BooleanVectorContents = UndefinedBooleanContent;
+ SchedPreferenceInfo = Sched::ILP;
JumpBufSize = 0;
JumpBufAlignment = 0;
MinFunctionAlignment = 0;
PrefLoopAlignment = 0;
MinStackArgumentAlignment = 1;
ShouldFoldAtomicFences = false;
+ InsertFencesForAtomic = false;
InitLibcallNames(LibcallRoutineNames);
InitCmpLibcallCCs(CmpLibcallCCs);
NewVT = EltTy;
IntermediateVT = NewVT;
+ unsigned NewVTSize = NewVT.getSizeInBits();
+
+ // Convert sizes such as i33 to i64.
+ if (!isPowerOf2_32(NewVTSize))
+ NewVTSize = NextPowerOf2(NewVTSize);
+
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());
+ return NumVectorRegs*(NewVTSize/DestVT.getSizeInBits());
// Otherwise, promotion or legal types use the same number of registers as
// the vector decimated to the appropriate level.
unsigned NElts = VT.getVectorNumElements();
if (NElts != 1) {
bool IsLegalWiderType = false;
+ // If we allow the promotion of vector elements using a flag,
+ // then return TypePromoteInteger on vector elements.
+ // First try to promote the elements of integer vectors. If no legal
+ // promotion was found, fallback to the widen-vector method.
+ if (mayPromoteElements)
for (unsigned nVT = i+1; nVT <= MVT::LAST_VECTOR_VALUETYPE; ++nVT) {
EVT SVT = (MVT::SimpleValueType)nVT;
-
- // If we allow the promotion of vector elements using a flag,
- // then return TypePromoteInteger on vector elements.
- if (mayPromoteElements) {
- // Promote vectors of integers to vectors with the same number
- // of elements, with a wider element type.
- if (SVT.getVectorElementType().getSizeInBits() > EltVT.getSizeInBits()
- && SVT.getVectorNumElements() == NElts &&
- isTypeLegal(SVT) && SVT.getScalarType().isInteger()) {
- TransformToType[i] = SVT;
- RegisterTypeForVT[i] = SVT;
- NumRegistersForVT[i] = 1;
- ValueTypeActions.setTypeAction(VT, TypePromoteInteger);
- IsLegalWiderType = true;
- break;
- }
+ // Promote vectors of integers to vectors with the same number
+ // of elements, with a wider element type.
+ if (SVT.getVectorElementType().getSizeInBits() > EltVT.getSizeInBits()
+ && SVT.getVectorNumElements() == NElts &&
+ isTypeLegal(SVT) && SVT.getScalarType().isInteger()) {
+ TransformToType[i] = SVT;
+ RegisterTypeForVT[i] = SVT;
+ NumRegistersForVT[i] = 1;
+ ValueTypeActions.setTypeAction(VT, TypePromoteInteger);
+ IsLegalWiderType = true;
+ break;
}
+ }
+
+ if (IsLegalWiderType) continue;
+ // Try to widen the vector.
+ for (unsigned nVT = i+1; nVT <= MVT::LAST_VECTOR_VALUETYPE; ++nVT) {
+ EVT SVT = (MVT::SimpleValueType)nVT;
if (SVT.getVectorElementType() == EltVT &&
SVT.getVectorNumElements() > NElts &&
isTypeLegal(SVT)) {
}
-MVT::SimpleValueType TargetLowering::getSetCCResultType(EVT VT) const {
+EVT TargetLowering::getSetCCResultType(EVT VT) const {
+ assert(!VT.isVector() && "No default SetCC type for vectors!");
return PointerTy.SimpleTy;
}
EVT DestVT = getRegisterType(Context, NewVT);
RegisterVT = DestVT;
+ unsigned NewVTSize = NewVT.getSizeInBits();
+
+ // Convert sizes such as i33 to i64.
+ if (!isPowerOf2_32(NewVTSize))
+ NewVTSize = NextPowerOf2(NewVTSize);
+
if (DestVT.bitsLT(NewVT)) // Value is expanded, e.g. i64 -> i16.
- return NumVectorRegs*(NewVT.getSizeInBits()/DestVT.getSizeInBits());
+ return NumVectorRegs*(NewVTSize/DestVT.getSizeInBits());
// Otherwise, promotion or legal types use the same number of registers as
// the vector decimated to the appropriate level.
/// type of the given function. This does not require a DAG or a return value,
/// and is suitable for use before any DAGs for the function are constructed.
/// TODO: Move this out of TargetLowering.cpp.
-void llvm::GetReturnInfo(const Type* ReturnType, Attributes attr,
+void llvm::GetReturnInfo(Type* ReturnType, Attributes attr,
SmallVectorImpl<ISD::OutputArg> &Outs,
const TargetLowering &TLI,
SmallVectorImpl<uint64_t> *Offsets) {
/// getByValTypeAlignment - Return the desired alignment for ByVal aggregate
/// function arguments in the caller parameter area. This is the actual
/// alignment, not its logarithm.
-unsigned TargetLowering::getByValTypeAlignment(const Type *Ty) const {
+unsigned TargetLowering::getByValTypeAlignment(Type *Ty) const {
return TD->getCallFrameTypeAlignment(Ty);
}
break;
}
case ISD::AssertZext: {
- // Demand all the bits of the input that are demanded in the output.
- // The low bits are obvious; the high bits are demanded because we're
- // asserting that they're zero here.
- if (SimplifyDemandedBits(Op.getOperand(0), NewMask,
+ // AssertZext demands all of the high bits, plus any of the low bits
+ // demanded by its users.
+ EVT VT = cast<VTSDNode>(Op.getOperand(1))->getVT();
+ APInt InMask = APInt::getLowBitsSet(BitWidth,
+ VT.getSizeInBits());
+ if (SimplifyDemandedBits(Op.getOperand(0), ~InMask | NewMask,
KnownZero, KnownOne, TLO, Depth+1))
return true;
assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
- EVT VT = cast<VTSDNode>(Op.getOperand(1))->getVT();
- APInt InMask = APInt::getLowBitsSet(BitWidth,
- VT.getSizeInBits());
KnownZero |= ~InMask & NewMask;
break;
}
case ISD::BITCAST:
- // 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 == APInt::getSignBit(Op.getValueType().getSizeInBits()) &&
- Op.getOperand(0).getValueType().isFloatingPoint() &&
- !Op.getOperand(0).getValueType().isVector()) {
- if (isOperationLegalOrCustom(ISD::FGETSIGN, MVT::i32)) {
- EVT Ty = (isOperationLegalOrCustom(ISD::FGETSIGN, Op.getValueType())) ?
- Op.getValueType() : MVT::i32;
+ // If this is an FP->Int bitcast and if the sign bit is the only
+ // thing demanded, turn this into a FGETSIGN.
+ if (!Op.getOperand(0).getValueType().isVector() &&
+ NewMask == APInt::getSignBit(Op.getValueType().getSizeInBits()) &&
+ Op.getOperand(0).getValueType().isFloatingPoint()) {
+ bool OpVTLegal = isOperationLegalOrCustom(ISD::FGETSIGN, Op.getValueType());
+ bool i32Legal = isOperationLegalOrCustom(ISD::FGETSIGN, MVT::i32);
+ if ((OpVTLegal || i32Legal) && Op.getValueType().isSimple()) {
+ EVT Ty = OpVTLegal ? Op.getValueType() : MVT::i32;
// Make a FGETSIGN + SHL to move the sign bit into the appropriate
// place. We expect the SHL to be eliminated by other optimizations.
SDValue Sign = TLO.DAG.getNode(ISD::FGETSIGN, dl, Ty, Op.getOperand(0));
- if (Ty != Op.getValueType())
+ unsigned OpVTSizeInBits = Op.getValueType().getSizeInBits();
+ if (!OpVTLegal && OpVTSizeInBits > 32)
Sign = TLO.DAG.getNode(ISD::ZERO_EXTEND, dl, Op.getValueType(), Sign);
unsigned ShVal = Op.getValueType().getSizeInBits()-1;
SDValue ShAmt = TLO.DAG.getConstant(ShVal, Op.getValueType());
// comparisons.
if (isa<ConstantSDNode>(N0.getNode()))
return DAG.getSetCC(dl, VT, N1, N0, ISD::getSetCCSwappedOperands(Cond));
-
+
if (ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1.getNode())) {
const APInt &C1 = N1C->getAPIntValue();
}
} else if (N1C->getAPIntValue() == 1 &&
(VT == MVT::i1 ||
- getBooleanContents() == ZeroOrOneBooleanContent)) {
+ getBooleanContents(false) == ZeroOrOneBooleanContent)) {
SDValue Op0 = N0;
if (Op0.getOpcode() == ISD::TRUNCATE)
Op0 = Op0.getOperand(0);
TargetLowering::ConstraintType
TargetLowering::getConstraintType(const std::string &Constraint) const {
- // FIXME: lots more standard ones to handle.
if (Constraint.size() == 1) {
switch (Constraint[0]) {
default: break;
/// LowerAsmOperandForConstraint - Lower the specified operand into the Ops
/// vector. If it is invalid, don't add anything to Ops.
void TargetLowering::LowerAsmOperandForConstraint(SDValue Op,
- char ConstraintLetter,
+ std::string &Constraint,
std::vector<SDValue> &Ops,
SelectionDAG &DAG) const {
+
+ if (Constraint.length() > 1) return;
+
+ char ConstraintLetter = Constraint[0];
switch (ConstraintLetter) {
default: break;
case 'X': // Allows any operand; labels (basic block) use this.
}
}
-std::vector<unsigned> TargetLowering::
-getRegClassForInlineAsmConstraint(const std::string &Constraint,
- EVT VT) const {
- return std::vector<unsigned>();
-}
-
-
std::pair<unsigned, const TargetRegisterClass*> TargetLowering::
getRegForInlineAsmConstraint(const std::string &Constraint,
EVT VT) const {
// 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();
- I != E; ++I) {
- if (isTypeLegal(*I)) {
- isLegal = true;
- break;
- }
- }
-
- if (!isLegal) continue;
+ if (!isLegalRC(RC))
+ continue;
for (TargetRegisterClass::iterator I = RC->begin(), E = RC->end();
I != E; ++I) {
// corresponding argument.
assert(!CS.getType()->isVoidTy() &&
"Bad inline asm!");
- if (const StructType *STy = dyn_cast<StructType>(CS.getType())) {
+ if (StructType *STy = dyn_cast<StructType>(CS.getType())) {
OpInfo.ConstraintVT = getValueType(STy->getElementType(ResNo));
} else {
assert(ResNo == 0 && "Asm only has one result!");
}
if (OpInfo.CallOperandVal) {
- const llvm::Type *OpTy = OpInfo.CallOperandVal->getType();
+ llvm::Type *OpTy = OpInfo.CallOperandVal->getType();
if (OpInfo.isIndirect) {
-
const llvm::PointerType *PtrTy = dyn_cast<PointerType>(OpTy);
+ llvm::PointerType *PtrTy = dyn_cast<PointerType>(OpTy);
if (!PtrTy)
report_fatal_error("Indirect operand for inline asm not a pointer!");
OpTy = PtrTy->getElementType();
}
-
+
// Look for vector wrapped in a struct. e.g. { <16 x i8> }.
- if (const StructType *STy = dyn_cast<StructType>(OpTy))
+ if (StructType *STy = dyn_cast<StructType>(OpTy))
if (STy->getNumElements() == 1)
OpTy = STy->getElementType(0);
AsmOperandInfo &Input = ConstraintOperands[OpInfo.MatchingInput];
if (OpInfo.ConstraintVT != Input.ConstraintVT) {
+ std::pair<unsigned, const TargetRegisterClass*> MatchRC =
+ getRegForInlineAsmConstraint(OpInfo.ConstraintCode, OpInfo.ConstraintVT);
+ std::pair<unsigned, const TargetRegisterClass*> InputRC =
+ getRegForInlineAsmConstraint(Input.ConstraintCode, Input.ConstraintVT);
if ((OpInfo.ConstraintVT.isInteger() !=
Input.ConstraintVT.isInteger()) ||
- (OpInfo.ConstraintVT.getSizeInBits() !=
- Input.ConstraintVT.getSizeInBits())) {
+ (MatchRC.second != InputRC.second)) {
report_fatal_error("Unsupported asm: input constraint"
" with a matching output constraint of"
" incompatible type!");
assert(OpInfo.Codes[i].size() == 1 &&
"Unhandled multi-letter 'other' constraint");
std::vector<SDValue> ResultOps;
- TLI.LowerAsmOperandForConstraint(Op, OpInfo.Codes[i][0],
+ TLI.LowerAsmOperandForConstraint(Op, OpInfo.Codes[i],
ResultOps, *DAG);
if (!ResultOps.empty()) {
BestType = CType;
/// isLegalAddressingMode - Return true if the addressing mode represented
/// by AM is legal for this target, for a load/store of the specified type.
bool TargetLowering::isLegalAddressingMode(const AddrMode &AM,
- const Type *Ty) const {
+ Type *Ty) const {
// The default implementation of this implements a conservative RISCy, r+r and
// r+i addr mode.
return true;
}
+/// BuildExactDiv - Given an exact SDIV by a constant, create a multiplication
+/// with the multiplicative inverse of the constant.
+SDValue TargetLowering::BuildExactSDIV(SDValue Op1, SDValue Op2, DebugLoc dl,
+ SelectionDAG &DAG) const {
+ ConstantSDNode *C = cast<ConstantSDNode>(Op2);
+ APInt d = C->getAPIntValue();
+ assert(d != 0 && "Division by zero!");
+
+ // Shift the value upfront if it is even, so the LSB is one.
+ unsigned ShAmt = d.countTrailingZeros();
+ if (ShAmt) {
+ // TODO: For UDIV use SRL instead of SRA.
+ SDValue Amt = DAG.getConstant(ShAmt, getShiftAmountTy(Op1.getValueType()));
+ Op1 = DAG.getNode(ISD::SRA, dl, Op1.getValueType(), Op1, Amt);
+ d = d.ashr(ShAmt);
+ }
+
+ // Calculate the multiplicative inverse, using Newton's method.
+ APInt t, xn = d;
+ while ((t = d*xn) != 1)
+ xn *= APInt(d.getBitWidth(), 2) - t;
+
+ Op2 = DAG.getConstant(xn, Op1.getValueType());
+ return DAG.getNode(ISD::MUL, dl, Op1.getValueType(), Op1, Op2);
+}
+
/// BuildSDIVSequence - Given an ISD::SDIV node expressing a divide by constant,
/// return a DAG expression to select that will generate the same value by
/// multiplying by a magic number. See:
/// <http://the.wall.riscom.net/books/proc/ppc/cwg/code2.html>
-SDValue TargetLowering::BuildSDIV(SDNode *N, SelectionDAG &DAG,
- std::vector<SDNode*>* Created) const {
+SDValue TargetLowering::
+BuildSDIV(SDNode *N, SelectionDAG &DAG, bool IsAfterLegalization,
+ std::vector<SDNode*>* Created) const {
EVT VT = N->getValueType(0);
DebugLoc dl= N->getDebugLoc();
// Multiply the numerator (operand 0) by the magic value
// FIXME: We should support doing a MUL in a wider type
SDValue Q;
- if (isOperationLegalOrCustom(ISD::MULHS, VT))
+ if (IsAfterLegalization ? isOperationLegal(ISD::MULHS, VT) :
+ isOperationLegalOrCustom(ISD::MULHS, VT))
Q = DAG.getNode(ISD::MULHS, dl, VT, N->getOperand(0),
DAG.getConstant(magics.m, VT));
- else if (isOperationLegalOrCustom(ISD::SMUL_LOHI, VT))
+ else if (IsAfterLegalization ? isOperationLegal(ISD::SMUL_LOHI, VT) :
+ isOperationLegalOrCustom(ISD::SMUL_LOHI, VT))
Q = SDValue(DAG.getNode(ISD::SMUL_LOHI, dl, DAG.getVTList(VT, VT),
N->getOperand(0),
DAG.getConstant(magics.m, VT)).getNode(), 1);
/// return a DAG expression to select that will generate the same value by
/// multiplying by a magic number. See:
/// <http://the.wall.riscom.net/books/proc/ppc/cwg/code2.html>
-SDValue TargetLowering::BuildUDIV(SDNode *N, SelectionDAG &DAG,
- std::vector<SDNode*>* Created) const {
+SDValue TargetLowering::
+BuildUDIV(SDNode *N, SelectionDAG &DAG, bool IsAfterLegalization,
+ std::vector<SDNode*>* Created) const {
EVT VT = N->getValueType(0);
DebugLoc dl = N->getDebugLoc();
// Multiply the numerator (operand 0) by the magic value
// FIXME: We should support doing a MUL in a wider type
- if (isOperationLegalOrCustom(ISD::MULHU, VT))
+ if (IsAfterLegalization ? isOperationLegal(ISD::MULHU, VT) :
+ isOperationLegalOrCustom(ISD::MULHU, VT))
Q = DAG.getNode(ISD::MULHU, dl, VT, Q, DAG.getConstant(magics.m, VT));
- else if (isOperationLegalOrCustom(ISD::UMUL_LOHI, VT))
+ else if (IsAfterLegalization ? isOperationLegal(ISD::UMUL_LOHI, VT) :
+ isOperationLegalOrCustom(ISD::UMUL_LOHI, VT))
Q = SDValue(DAG.getNode(ISD::UMUL_LOHI, dl, DAG.getVTList(VT, VT), Q,
DAG.getConstant(magics.m, VT)).getNode(), 1);
else
projects / oota-llvm.git / blobdiff