#include "llvm/Target/TargetLowering.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/Target/TargetRegisterInfo.h"
-#include "llvm/Target/TargetSubtarget.h"
#include "llvm/GlobalVariable.h"
#include "llvm/DerivedTypes.h"
+#include "llvm/CodeGen/Analysis.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineJumpTableInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/Support/ErrorHandling.h"
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::FPROUND_F80_F64] = "__truncxfdf2";
Names[RTLIB::FPROUND_PPCF128_F64] = "__trunctfdf2";
- Names[RTLIB::FPTOSINT_F32_I8] = "__fixsfi8";
- Names[RTLIB::FPTOSINT_F32_I16] = "__fixsfi16";
+ Names[RTLIB::FPTOSINT_F32_I8] = "__fixsfqi";
+ Names[RTLIB::FPTOSINT_F32_I16] = "__fixsfhi";
Names[RTLIB::FPTOSINT_F32_I32] = "__fixsfsi";
Names[RTLIB::FPTOSINT_F32_I64] = "__fixsfdi";
Names[RTLIB::FPTOSINT_F32_I128] = "__fixsfti";
+ Names[RTLIB::FPTOSINT_F64_I8] = "__fixdfqi";
+ Names[RTLIB::FPTOSINT_F64_I16] = "__fixdfhi";
Names[RTLIB::FPTOSINT_F64_I32] = "__fixdfsi";
Names[RTLIB::FPTOSINT_F64_I64] = "__fixdfdi";
Names[RTLIB::FPTOSINT_F64_I128] = "__fixdfti";
Names[RTLIB::FPTOSINT_PPCF128_I32] = "__fixtfsi";
Names[RTLIB::FPTOSINT_PPCF128_I64] = "__fixtfdi";
Names[RTLIB::FPTOSINT_PPCF128_I128] = "__fixtfti";
- Names[RTLIB::FPTOUINT_F32_I8] = "__fixunssfi8";
- Names[RTLIB::FPTOUINT_F32_I16] = "__fixunssfi16";
+ Names[RTLIB::FPTOUINT_F32_I8] = "__fixunssfqi";
+ Names[RTLIB::FPTOUINT_F32_I16] = "__fixunssfhi";
Names[RTLIB::FPTOUINT_F32_I32] = "__fixunssfsi";
Names[RTLIB::FPTOUINT_F32_I64] = "__fixunssfdi";
Names[RTLIB::FPTOUINT_F32_I128] = "__fixunssfti";
+ Names[RTLIB::FPTOUINT_F64_I8] = "__fixunsdfqi";
+ Names[RTLIB::FPTOUINT_F64_I16] = "__fixunsdfhi";
Names[RTLIB::FPTOUINT_F64_I32] = "__fixunsdfsi";
Names[RTLIB::FPTOUINT_F64_I64] = "__fixunsdfdi";
Names[RTLIB::FPTOUINT_F64_I128] = "__fixunsdfti";
Names[RTLIB::MEMMOVE] = "memmove";
Names[RTLIB::MEMSET] = "memset";
Names[RTLIB::UNWIND_RESUME] = "_Unwind_Resume";
+ Names[RTLIB::SYNC_VAL_COMPARE_AND_SWAP_1] = "__sync_val_compare_and_swap_1";
+ Names[RTLIB::SYNC_VAL_COMPARE_AND_SWAP_2] = "__sync_val_compare_and_swap_2";
+ Names[RTLIB::SYNC_VAL_COMPARE_AND_SWAP_4] = "__sync_val_compare_and_swap_4";
+ Names[RTLIB::SYNC_VAL_COMPARE_AND_SWAP_8] = "__sync_val_compare_and_swap_8";
+ Names[RTLIB::SYNC_LOCK_TEST_AND_SET_1] = "__sync_lock_test_and_set_1";
+ Names[RTLIB::SYNC_LOCK_TEST_AND_SET_2] = "__sync_lock_test_and_set_2";
+ Names[RTLIB::SYNC_LOCK_TEST_AND_SET_4] = "__sync_lock_test_and_set_4";
+ Names[RTLIB::SYNC_LOCK_TEST_AND_SET_8] = "__sync_lock_test_and_set_8";
+ Names[RTLIB::SYNC_FETCH_AND_ADD_1] = "__sync_fetch_and_add_1";
+ Names[RTLIB::SYNC_FETCH_AND_ADD_2] = "__sync_fetch_and_add_2";
+ Names[RTLIB::SYNC_FETCH_AND_ADD_4] = "__sync_fetch_and_add_4";
+ Names[RTLIB::SYNC_FETCH_AND_ADD_8] = "__sync_fetch_and_add_8";
+ Names[RTLIB::SYNC_FETCH_AND_SUB_1] = "__sync_fetch_and_sub_1";
+ Names[RTLIB::SYNC_FETCH_AND_SUB_2] = "__sync_fetch_and_sub_2";
+ Names[RTLIB::SYNC_FETCH_AND_SUB_4] = "__sync_fetch_and_sub_4";
+ Names[RTLIB::SYNC_FETCH_AND_SUB_8] = "__sync_fetch_and_sub_8";
+ Names[RTLIB::SYNC_FETCH_AND_AND_1] = "__sync_fetch_and_and_1";
+ Names[RTLIB::SYNC_FETCH_AND_AND_2] = "__sync_fetch_and_and_2";
+ Names[RTLIB::SYNC_FETCH_AND_AND_4] = "__sync_fetch_and_and_4";
+ Names[RTLIB::SYNC_FETCH_AND_AND_8] = "__sync_fetch_and_and_8";
+ Names[RTLIB::SYNC_FETCH_AND_OR_1] = "__sync_fetch_and_or_1";
+ Names[RTLIB::SYNC_FETCH_AND_OR_2] = "__sync_fetch_and_or_2";
+ Names[RTLIB::SYNC_FETCH_AND_OR_4] = "__sync_fetch_and_or_4";
+ 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_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";
+ Names[RTLIB::SYNC_FETCH_AND_NAND_4] = "__sync_fetch_and_nand_4";
+ Names[RTLIB::SYNC_FETCH_AND_NAND_8] = "__sync_fetch_and_nand_8";
}
/// InitLibcallCallingConvs - Set default libcall CallingConvs.
if (RetVT == MVT::f64)
return FPEXT_F32_F64;
}
+
return UNKNOWN_LIBCALL;
}
if (OpVT == MVT::ppcf128)
return FPROUND_PPCF128_F64;
}
+
return UNKNOWN_LIBCALL;
}
if (RetVT == MVT::i128)
return FPTOSINT_F32_I128;
} else if (OpVT == MVT::f64) {
+ if (RetVT == MVT::i8)
+ return FPTOSINT_F64_I8;
+ if (RetVT == MVT::i16)
+ return FPTOSINT_F64_I16;
if (RetVT == MVT::i32)
return FPTOSINT_F64_I32;
if (RetVT == MVT::i64)
if (RetVT == MVT::i128)
return FPTOUINT_F32_I128;
} else if (OpVT == MVT::f64) {
+ if (RetVT == MVT::i8)
+ return FPTOUINT_F64_I8;
+ if (RetVT == MVT::i16)
+ return FPTOUINT_F64_I16;
if (RetVT == MVT::i32)
return FPTOUINT_F64_I32;
if (RetVT == MVT::i64)
}
/// NOTE: The constructor takes ownership of TLOF.
-TargetLowering::TargetLowering(TargetMachine &tm,TargetLoweringObjectFile *tlof)
+TargetLowering::TargetLowering(const TargetMachine &tm,
+ const TargetLoweringObjectFile *tlof)
: TM(tm), TD(TM.getTargetData()), TLOF(*tlof) {
// All operations default to being supported.
memset(OpActions, 0, sizeof(OpActions));
memset(LoadExtActions, 0, sizeof(LoadExtActions));
memset(TruncStoreActions, 0, sizeof(TruncStoreActions));
memset(IndexedModeActions, 0, sizeof(IndexedModeActions));
- memset(ConvertActions, 0, sizeof(ConvertActions));
memset(CondCodeActions, 0, sizeof(CondCodeActions));
// Set default actions for various operations.
setOperationAction(ISD::TRAP, MVT::Other, Expand);
IsLittleEndian = TD->isLittleEndian();
- UsesGlobalOffsetTable = false;
ShiftAmountTy = PointerTy = MVT::getIntegerVT(8*TD->getPointerSize());
memset(RegClassForVT, 0,MVT::LAST_VALUETYPE*sizeof(TargetRegisterClass*));
memset(TargetDAGCombineArray, 0, array_lengthof(TargetDAGCombineArray));
ExceptionPointerRegister = 0;
ExceptionSelectorRegister = 0;
BooleanContents = UndefinedBooleanContent;
- SchedPreferenceInfo = SchedulingForLatency;
+ SchedPreferenceInfo = Sched::Latency;
JumpBufSize = 0;
JumpBufAlignment = 0;
- IfCvtBlockSizeLimit = 2;
- IfCvtDupBlockSizeLimit = 0;
PrefLoopAlignment = 0;
+ MinStackArgumentAlignment = 1;
+ ShouldFoldAtomicFences = false;
InitLibcallNames(LibcallRoutineNames);
InitCmpLibcallCCs(CmpLibcallCCs);
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) {
+ unsigned &NumIntermediates,
+ EVT &RegisterVT,
+ TargetLowering *TLI) {
// Figure out the right, legal destination reg to copy into.
unsigned NumElts = VT.getVectorNumElements();
MVT EltTy = VT.getVectorElementType();
EVT DestVT = TLI->getRegisterType(NewVT);
RegisterVT = DestVT;
- if (EVT(DestVT).bitsLT(NewVT)) {
- // Value is expanded, e.g. i64 -> i16.
+ 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;
+ // Otherwise, promotion or legal types use the same number of registers as
+ // the vector decimated to the appropriate level.
+ return NumVectorRegs;
+}
+
+/// isLegalRC - Return true if the value types that can be represented by the
+/// specified register class are all legal.
+bool TargetLowering::isLegalRC(const TargetRegisterClass *RC) const {
+ for (TargetRegisterClass::vt_iterator I = RC->vt_begin(), E = RC->vt_end();
+ I != E; ++I) {
+ if (isTypeLegal(*I))
+ return true;
+ }
+ return false;
+}
+
+/// hasLegalSuperRegRegClasses - Return true if the specified register class
+/// has one or more super-reg register classes that are legal.
+bool
+TargetLowering::hasLegalSuperRegRegClasses(const TargetRegisterClass *RC) const{
+ if (*RC->superregclasses_begin() == 0)
+ return false;
+ for (TargetRegisterInfo::regclass_iterator I = RC->superregclasses_begin(),
+ E = RC->superregclasses_end(); I != E; ++I) {
+ const TargetRegisterClass *RRC = *I;
+ if (isLegalRC(RRC))
+ return true;
+ }
+ return false;
}
+/// findRepresentativeClass - Return the largest legal super-reg register class
+/// of the register class for the specified type and its associated "cost".
+std::pair<const TargetRegisterClass*, uint8_t>
+TargetLowering::findRepresentativeClass(EVT VT) const {
+ const TargetRegisterClass *RC = RegClassForVT[VT.getSimpleVT().SimpleTy];
+ if (!RC)
+ return std::make_pair(RC, 0);
+ const TargetRegisterClass *BestRC = RC;
+ for (TargetRegisterInfo::regclass_iterator I = RC->superregclasses_begin(),
+ E = RC->superregclasses_end(); I != E; ++I) {
+ const TargetRegisterClass *RRC = *I;
+ if (RRC->isASubClass() || !isLegalRC(RRC))
+ continue;
+ if (!hasLegalSuperRegRegClasses(RRC))
+ return std::make_pair(RRC, 1);
+ BestRC = RRC;
+ }
+ return std::make_pair(BestRC, 1);
+}
+
+
/// computeRegisterProperties - Once all of the register classes are added,
/// this allows us to compute derived properties we expose.
void TargetLowering::computeRegisterProperties() {
for (unsigned i = MVT::FIRST_VECTOR_VALUETYPE;
i <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++i) {
MVT VT = (MVT::SimpleValueType)i;
- if (!isTypeLegal(VT)) {
- MVT IntermediateVT;
- EVT RegisterVT;
- unsigned NumIntermediates;
- NumRegistersForVT[i] =
- getVectorTypeBreakdownMVT(VT, IntermediateVT, NumIntermediates,
- RegisterVT, this);
- RegisterTypeForVT[i] = RegisterVT;
-
- // Determine if there is a legal wider type.
+ if (isTypeLegal(VT)) continue;
+
+ // Determine if there is a legal wider type. If so, we should promote to
+ // that wider vector type.
+ EVT EltVT = VT.getVectorElementType();
+ unsigned NElts = VT.getVectorNumElements();
+ if (NElts != 1) {
bool IsLegalWiderType = false;
- EVT EltVT = VT.getVectorElementType();
- unsigned NElts = VT.getVectorNumElements();
for (unsigned nVT = i+1; nVT <= MVT::LAST_VECTOR_VALUETYPE; ++nVT) {
EVT SVT = (MVT::SimpleValueType)nVT;
- if (isTypeLegal(SVT) && SVT.getVectorElementType() == EltVT &&
- SVT.getVectorNumElements() > NElts && NElts != 1) {
+ if (SVT.getVectorElementType() == EltVT &&
+ SVT.getVectorNumElements() > NElts &&
+ isTypeSynthesizable(SVT)) {
TransformToType[i] = SVT;
+ RegisterTypeForVT[i] = SVT;
+ NumRegistersForVT[i] = 1;
ValueTypeActions.setTypeAction(VT, Promote);
IsLegalWiderType = true;
break;
}
}
- if (!IsLegalWiderType) {
- EVT NVT = VT.getPow2VectorType();
- if (NVT == VT) {
- // Type is already a power of 2. The default action is to split.
- TransformToType[i] = MVT::Other;
- ValueTypeActions.setTypeAction(VT, Expand);
- } else {
- TransformToType[i] = NVT;
- ValueTypeActions.setTypeAction(VT, Promote);
- }
- }
+ if (IsLegalWiderType) continue;
}
+
+ MVT IntermediateVT;
+ EVT RegisterVT;
+ unsigned NumIntermediates;
+ NumRegistersForVT[i] =
+ getVectorTypeBreakdownMVT(VT, IntermediateVT, NumIntermediates,
+ RegisterVT, this);
+ RegisterTypeForVT[i] = RegisterVT;
+
+ EVT NVT = VT.getPow2VectorType();
+ if (NVT == VT) {
+ // Type is already a power of 2. The default action is to split.
+ TransformToType[i] = MVT::Other;
+ ValueTypeActions.setTypeAction(VT, Expand);
+ } else {
+ TransformToType[i] = NVT;
+ ValueTypeActions.setTypeAction(VT, Promote);
+ }
+ }
+
+ // Determine the 'representative' register class for each value type.
+ // An representative register class is the largest (meaning one which is
+ // not a sub-register class / subreg register class) legal register class for
+ // a group of value types. For example, on i386, i8, i16, and i32
+ // representative would be GR32; while on x86_64 it's GR64.
+ for (unsigned i = 0; i != MVT::LAST_VALUETYPE; ++i) {
+ const TargetRegisterClass* RRC;
+ uint8_t Cost;
+ tie(RRC, Cost) = findRepresentativeClass((MVT::SimpleValueType)i);
+ RepRegClassForVT[i] = RRC;
+ RepRegClassCostForVT[i] = Cost;
}
}
EVT &IntermediateVT,
unsigned &NumIntermediates,
EVT &RegisterVT) const {
- // Figure out the right, legal destination reg to copy into.
unsigned NumElts = VT.getVectorNumElements();
+
+ // If there is a wider vector type with the same element type as this one,
+ // we should widen to that legal vector type. This handles things like
+ // <2 x float> -> <4 x float>.
+ if (NumElts != 1 && getTypeAction(VT) == Promote) {
+ RegisterVT = getTypeToTransformTo(Context, VT);
+ if (isTypeLegal(RegisterVT)) {
+ IntermediateVT = RegisterVT;
+ NumIntermediates = 1;
+ return 1;
+ }
+ }
+
+ // Figure out the right, legal destination reg to copy into.
EVT EltTy = VT.getVectorElementType();
unsigned NumVectorRegs = 1;
EVT DestVT = getRegisterType(Context, NewVT);
RegisterVT = DestVT;
- if (DestVT.bitsLT(NewVT)) {
- // Value is expanded, e.g. i64 -> i16.
+ if (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;
+ // Otherwise, promotion or legal types use the same number of registers as
+ // the vector decimated to the appropriate level.
+ return NumVectorRegs;
}
-/// 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.
-EVT TargetLowering::getWidenVectorType(EVT VT) const {
- assert(VT.isVector());
- if (isTypeLegal(VT))
- return VT;
-
- // Default is not to widen until moved to LegalizeTypes
- return MVT::Other;
+/// Get the EVTs and ArgFlags collections that represent the legalized return
+/// 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,
+ SmallVectorImpl<ISD::OutputArg> &Outs,
+ const TargetLowering &TLI,
+ SmallVectorImpl<uint64_t> *Offsets) {
+ SmallVector<EVT, 4> ValueVTs;
+ ComputeValueVTs(TLI, ReturnType, ValueVTs);
+ unsigned NumValues = ValueVTs.size();
+ if (NumValues == 0) return;
+ unsigned Offset = 0;
+
+ for (unsigned j = 0, f = NumValues; j != f; ++j) {
+ EVT VT = ValueVTs[j];
+ ISD::NodeType ExtendKind = ISD::ANY_EXTEND;
+
+ if (attr & Attribute::SExt)
+ ExtendKind = ISD::SIGN_EXTEND;
+ else if (attr & Attribute::ZExt)
+ ExtendKind = ISD::ZERO_EXTEND;
+
+ // 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. The frontend should mark functions whose return values
+ // require promoting with signext or zeroext attributes.
+ if (ExtendKind != ISD::ANY_EXTEND && VT.isInteger()) {
+ EVT MinVT = TLI.getRegisterType(ReturnType->getContext(), MVT::i32);
+ if (VT.bitsLT(MinVT))
+ VT = MinVT;
+ }
+
+ unsigned NumParts = TLI.getNumRegisters(ReturnType->getContext(), VT);
+ EVT PartVT = TLI.getRegisterType(ReturnType->getContext(), VT);
+ unsigned PartSize = TLI.getTargetData()->getTypeAllocSize(
+ PartVT.getTypeForEVT(ReturnType->getContext()));
+
+ // 'inreg' on function refers to return value
+ ISD::ArgFlagsTy Flags = ISD::ArgFlagsTy();
+ if (attr & Attribute::InReg)
+ Flags.setInReg();
+
+ // Propagate extension type if any
+ if (attr & Attribute::SExt)
+ Flags.setSExt();
+ else if (attr & Attribute::ZExt)
+ Flags.setZExt();
+
+ for (unsigned i = 0; i < NumParts; ++i) {
+ Outs.push_back(ISD::OutputArg(Flags, PartVT, /*isFixed=*/true));
+ if (Offsets) {
+ Offsets->push_back(Offset);
+ Offset += PartSize;
+ }
+ }
+ }
}
/// getByValTypeAlignment - Return the desired alignment for ByVal aggregate
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 (TLO.ShrinkDemandedConstant(Op, ~KnownZero2 & NewMask))
return true;
// If the operation can be done in a smaller type, do so.
- if (TLO.ShrinkOps && TLO.ShrinkDemandedOp(Op, BitWidth, NewMask, dl))
+ if (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.ShrinkOps && TLO.ShrinkDemandedOp(Op, BitWidth, NewMask, dl))
+ if (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.ShrinkOps && TLO.ShrinkDemandedOp(Op, BitWidth, NewMask, dl))
+ if (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
}
}
- if (SimplifyDemandedBits(Op.getOperand(0), NewMask.lshr(ShAmt),
+ if (SimplifyDemandedBits(InOp, NewMask.lshr(ShAmt),
KnownZero, KnownOne, TLO, Depth+1))
return true;
+
+ // Convert (shl (anyext x, c)) to (anyext (shl x, c)) if the high bits
+ // are not demanded. This will likely allow the anyext to be folded away.
+ if (InOp.getNode()->getOpcode() == ISD::ANY_EXTEND) {
+ SDValue InnerOp = InOp.getNode()->getOperand(0);
+ EVT InnerVT = InnerOp.getValueType();
+ if ((APInt::getHighBitsSet(BitWidth,
+ BitWidth - InnerVT.getSizeInBits()) &
+ DemandedMask) == 0 &&
+ isTypeDesirableForOp(ISD::SHL, InnerVT)) {
+ EVT ShTy = getShiftAmountTy();
+ if (!APInt(BitWidth, ShAmt).isIntN(ShTy.getSizeInBits()))
+ ShTy = InnerVT;
+ SDValue NarrowShl =
+ TLO.DAG.getNode(ISD::SHL, dl, InnerVT, InnerOp,
+ TLO.DAG.getConstant(ShAmt, ShTy));
+ return
+ TLO.CombineTo(Op,
+ TLO.DAG.getNode(ISD::ANY_EXTEND, dl, Op.getValueType(),
+ NarrowShl));
+ }
+ }
+
KnownZero <<= SA->getZExtValue();
KnownOne <<= SA->getZExtValue();
// low bits known zero.
// variable. The low bit of the shift cannot be an input sign bit unless
// the shift amount is >= the size of the datatype, which is undefined.
if (DemandedMask == 1)
- return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::SRL, dl, Op.getValueType(),
- Op.getOperand(0), Op.getOperand(1)));
+ return TLO.CombineTo(Op,
+ TLO.DAG.getNode(ISD::SRL, dl, Op.getValueType(),
+ Op.getOperand(0), Op.getOperand(1)));
if (ConstantSDNode *SA = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
EVT VT = Op.getValueType();
// present in the input.
APInt NewBits =
APInt::getHighBitsSet(BitWidth,
- BitWidth - EVT.getScalarType().getSizeInBits()) &
- NewMask;
+ BitWidth - EVT.getScalarType().getSizeInBits());
// If none of the extended bits are demanded, eliminate the sextinreg.
- if (NewBits == 0)
+ if ((NewBits & NewMask) == 0)
return TLO.CombineTo(Op, Op.getOperand(0));
APInt InSignBit = APInt::getSignBit(EVT.getScalarType().getSizeInBits());
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);
- HighBits = HighBits.lshr(ShAmt->getZExtValue());
- HighBits.trunc(BitWidth);
-
- if (ShAmt->getZExtValue() < BitWidth && !(HighBits & NewMask)) {
- // None of the shifted in bits are needed. Add a truncate of the
- // shift input, then shift it.
- SDValue NewTrunc = TLO.DAG.getNode(ISD::TRUNCATE, dl,
- Op.getValueType(),
- In.getOperand(0));
- return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::SRL, dl,
- Op.getValueType(),
- NewTrunc,
- In.getOperand(1)));
- }
+ if (TLO.LegalTypes() &&
+ !isTypeDesirableForOp(ISD::SRL, Op.getValueType()))
+ // Do not turn (vt1 truncate (vt2 srl)) into (vt1 srl) if vt1 is
+ // undesirable.
+ break;
+ ConstantSDNode *ShAmt = dyn_cast<ConstantSDNode>(In.getOperand(1));
+ if (!ShAmt)
+ break;
+ APInt HighBits = APInt::getHighBitsSet(OperandBitWidth,
+ OperandBitWidth - BitWidth);
+ HighBits = HighBits.lshr(ShAmt->getZExtValue());
+ HighBits.trunc(BitWidth);
+
+ if (ShAmt->getZExtValue() < BitWidth && !(HighBits & NewMask)) {
+ // None of the shifted in bits are needed. Add a truncate of the
+ // shift input, then shift it.
+ SDValue NewTrunc = TLO.DAG.getNode(ISD::TRUNCATE, dl,
+ Op.getValueType(),
+ In.getOperand(0));
+ return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::SRL, dl,
+ Op.getValueType(),
+ NewTrunc,
+ In.getOperand(1)));
}
break;
}
break;
}
case ISD::AssertZext: {
- EVT VT = cast<VTSDNode>(Op.getOperand(1))->getVT();
- APInt InMask = APInt::getLowBitsSet(BitWidth,
- VT.getSizeInBits());
- if (SimplifyDemandedBits(Op.getOperand(0), InMask & NewMask,
+ // 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,
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;
}
KnownOne2, TLO, Depth+1))
return true;
// See if the operation should be performed at a smaller bit width.
- if (TLO.ShrinkOps && TLO.ShrinkDemandedOp(Op, BitWidth, NewMask, dl))
+ if (TLO.ShrinkDemandedOp(Op, BitWidth, NewMask, dl))
return true;
}
// FALL THROUGH
// Fall back to ComputeMaskedBits to catch other known cases.
EVT OpVT = Val.getValueType();
- unsigned BitWidth = OpVT.getSizeInBits();
+ unsigned BitWidth = OpVT.getScalarType().getSizeInBits();
APInt Mask = APInt::getAllOnesValue(BitWidth);
APInt KnownZero, KnownOne;
DAG.ComputeMaskedBits(Val, Mask, KnownZero, KnownOne);
DAG.getConstant(bestOffset, PtrType));
unsigned NewAlign = MinAlign(Lod->getAlignment(), bestOffset);
SDValue NewLoad = DAG.getLoad(newVT, dl, Lod->getChain(), Ptr,
- Lod->getSrcValue(),
- Lod->getSrcValueOffset() + bestOffset,
- false, NewAlign);
+ Lod->getPointerInfo().getWithOffset(bestOffset),
+ false, false, NewAlign);
return DAG.getSetCC(dl, VT,
DAG.getNode(ISD::AND, dl, newVT, NewLoad,
DAG.getConstant(bestMask.trunc(bestWidth),
break; // todo, be more careful with signed comparisons
}
} else if (N0.getOpcode() == ISD::SIGN_EXTEND_INREG &&
- (Cond == ISD::SETEQ || Cond == ISD::SETNE)) {
+ (Cond == ISD::SETEQ || Cond == ISD::SETNE)) {
EVT ExtSrcTy = cast<VTSDNode>(N0.getOperand(1))->getVT();
unsigned ExtSrcTyBits = ExtSrcTy.getSizeInBits();
EVT ExtDstTy = N0.getValueType();
unsigned ExtDstTyBits = ExtDstTy.getSizeInBits();
- // If the extended part has any inconsistent bits, it cannot ever
- // compare equal. In other words, they have to be all ones or all
- // zeros.
- APInt ExtBits =
- APInt::getHighBitsSet(ExtDstTyBits, ExtDstTyBits - ExtSrcTyBits);
- if ((C1 & ExtBits) != 0 && (C1 & ExtBits) != ExtBits)
+ // If the constant doesn't fit into the number of bits for the source of
+ // the sign extension, it is impossible for both sides to be equal.
+ if (C1.getMinSignedBits() > ExtSrcTyBits)
return DAG.getConstant(Cond == ISD::SETNE, VT);
SDValue ZextOp;
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) {
+ 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().bitsGT(VT))
+ Op0 = DAG.getNode(ISD::AND, dl, VT,
+ DAG.getNode(ISD::TRUNCATE, dl, VT, Op0.getOperand(0)),
+ DAG.getConstant(1, VT));
+ else if (Op0.getValueType().bitsLT(VT))
+ Op0 = DAG.getNode(ISD::AND, dl, VT,
+ DAG.getNode(ISD::ANY_EXTEND, 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);
+ }
}
}
/// isGAPlusOffset - Returns true (and the GlobalValue and the offset) if the
/// node is a GlobalAddress + offset.
-bool TargetLowering::isGAPlusOffset(SDNode *N, GlobalValue* &GA,
+bool TargetLowering::isGAPlusOffset(SDNode *N, const GlobalValue* &GA,
int64_t &Offset) const {
if (isa<GlobalAddressSDNode>(N)) {
GlobalAddressSDNode *GASD = cast<GlobalAddressSDNode>(N);
return C_Memory;
case 'i': // Simple Integer or Relocatable Constant
case 'n': // Simple Integer
+ case 'E': // Floating Point Constant
+ case 'F': // Floating Point Constant
case 's': // Relocatable Constant
+ case 'p': // Address.
case 'X': // Allow ANY value.
case 'I': // Target registers.
case 'J':
case 'N':
case 'O':
case 'P':
+ case '<':
+ case '>':
return C_Other;
}
}
/// vector. If it is invalid, don't add anything to Ops.
void TargetLowering::LowerAsmOperandForConstraint(SDValue Op,
char ConstraintLetter,
- bool hasMemory,
std::vector<SDValue> &Ops,
SelectionDAG &DAG) const {
switch (ConstraintLetter) {
if (ConstraintLetter != 'n') {
int64_t Offs = GA->getOffset();
if (C) Offs += C->getZExtValue();
- Ops.push_back(DAG.getTargetGlobalAddress(GA->getGlobal(),
+ Ops.push_back(DAG.getTargetGlobalAddress(GA->getGlobal(),
+ C ? C->getDebugLoc() : DebugLoc(),
Op.getValueType(), Offs));
return;
}
getRegForInlineAsmConstraint(const std::string &Constraint,
EVT VT) const {
if (Constraint[0] != '{')
- return std::pair<unsigned, const TargetRegisterClass*>(0, 0);
+ return std::make_pair(0u, static_cast<TargetRegisterClass*>(0));
assert(*(Constraint.end()-1) == '}' && "Not a brace enclosed constraint?");
// Remove the braces from around the name.
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();
}
}
- return std::pair<unsigned, const TargetRegisterClass*>(0, 0);
+ return std::make_pair(0u, static_cast<const TargetRegisterClass*>(0));
}
//===----------------------------------------------------------------------===//
return atoi(ConstraintCode.c_str());
}
+
+/// ParseConstraints - Split up the constraint string from the inline
+/// assembly value into the specific constraints and their prefixes,
+/// and also tie in the associated operand values.
+/// If this returns an empty vector, and if the constraint string itself
+/// isn't empty, there was an error parsing.
+std::vector<TargetLowering::AsmOperandInfo> TargetLowering::ParseConstraints(
+ ImmutableCallSite CS) const {
+ /// ConstraintOperands - Information about all of the constraints.
+ std::vector<AsmOperandInfo> ConstraintOperands;
+ const InlineAsm *IA = cast<InlineAsm>(CS.getCalledValue());
+ unsigned maCount = 0; // Largest number of multiple alternative constraints.
+
+ // Do a prepass over the constraints, canonicalizing them, and building up the
+ // ConstraintOperands list.
+ std::vector<InlineAsm::ConstraintInfo>
+ ConstraintInfos = IA->ParseConstraints();
+
+ 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(AsmOperandInfo(ConstraintInfos[i]));
+ AsmOperandInfo &OpInfo = ConstraintOperands.back();
+
+ // Update multiple alternative constraint count.
+ if (OpInfo.multipleAlternatives.size() > maCount)
+ maCount = OpInfo.multipleAlternatives.size();
+
+ EVT OpVT = MVT::Other;
+
+ // Compute the value type for each operand.
+ switch (OpInfo.Type) {
+ case InlineAsm::isOutput:
+ // Indirect outputs just consume an argument.
+ if (OpInfo.isIndirect) {
+ OpInfo.CallOperandVal = const_cast<Value *>(CS.getArgument(ArgNo++));
+ break;
+ }
+
+ // The return value of the call is this value. As such, there is no
+ // corresponding argument.
+ assert(!CS.getType()->isVoidTy() &&
+ "Bad inline asm!");
+ if (const StructType *STy = dyn_cast<StructType>(CS.getType())) {
+ OpVT = getValueType(STy->getElementType(ResNo));
+ } else {
+ assert(ResNo == 0 && "Asm only has one result!");
+ OpVT = getValueType(CS.getType());
+ }
+ ++ResNo;
+ break;
+ case InlineAsm::isInput:
+ OpInfo.CallOperandVal = const_cast<Value *>(CS.getArgument(ArgNo++));
+ break;
+ case InlineAsm::isClobber:
+ // Nothing to do.
+ break;
+ }
+ }
+
+ // If we have multiple alternative constraints, select the best alternative.
+ if (ConstraintInfos.size()) {
+ if (maCount) {
+ unsigned bestMAIndex = 0;
+ int bestWeight = -1;
+ // weight: -1 = invalid match, and 0 = so-so match to 5 = good match.
+ int weight = -1;
+ unsigned maIndex;
+ // Compute the sums of the weights for each alternative, keeping track
+ // of the best (highest weight) one so far.
+ for (maIndex = 0; maIndex < maCount; ++maIndex) {
+ int weightSum = 0;
+ for (unsigned cIndex = 0, eIndex = ConstraintOperands.size();
+ cIndex != eIndex; ++cIndex) {
+ AsmOperandInfo& OpInfo = ConstraintOperands[cIndex];
+ if (OpInfo.Type == InlineAsm::isClobber)
+ continue;
+
+ // 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
+ // maCantMatch.
+ if (OpInfo.hasMatchingInput()) {
+ AsmOperandInfo &Input = ConstraintOperands[OpInfo.MatchingInput];
+
+ if (OpInfo.ConstraintVT != Input.ConstraintVT) {
+ if ((OpInfo.ConstraintVT.isInteger() !=
+ Input.ConstraintVT.isInteger()) ||
+ (OpInfo.ConstraintVT.getSizeInBits() !=
+ Input.ConstraintVT.getSizeInBits())) {
+ weightSum = -1; // Can't match.
+ break;
+ }
+ Input.ConstraintVT = OpInfo.ConstraintVT;
+ }
+ }
+
+ weight = getMultipleConstraintMatchWeight(OpInfo, maIndex);
+ if (weight == -1) {
+ weightSum = -1;
+ break;
+ }
+ weightSum += weight;
+ }
+ // Update best.
+ if (weightSum > bestWeight) {
+ bestWeight = weightSum;
+ bestMAIndex = maIndex;
+ }
+ }
+
+ // Now select chosen alternative in each constraint.
+ for (unsigned cIndex = 0, eIndex = ConstraintOperands.size();
+ cIndex != eIndex; ++cIndex) {
+ AsmOperandInfo& cInfo = ConstraintOperands[cIndex];
+ if (cInfo.Type == InlineAsm::isClobber)
+ continue;
+ cInfo.selectAlternative(bestMAIndex);
+ }
+ }
+ }
+
+ // Check and hook up tied operands, choose constraint code to use.
+ for (unsigned cIndex = 0, eIndex = ConstraintOperands.size();
+ cIndex != eIndex; ++cIndex) {
+ AsmOperandInfo& OpInfo = ConstraintOperands[cIndex];
+
+ // 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()) {
+ AsmOperandInfo &Input = ConstraintOperands[OpInfo.MatchingInput];
+
+ if (OpInfo.ConstraintVT != Input.ConstraintVT) {
+ if ((OpInfo.ConstraintVT.isInteger() !=
+ Input.ConstraintVT.isInteger()) ||
+ (OpInfo.ConstraintVT.getSizeInBits() !=
+ Input.ConstraintVT.getSizeInBits())) {
+ report_fatal_error("Unsupported asm: input constraint"
+ " with a matching output constraint of"
+ " incompatible type!");
+ }
+ Input.ConstraintVT = OpInfo.ConstraintVT;
+ }
+ }
+ }
+
+ return ConstraintOperands;
+}
+
/// getConstraintGenerality - Return an integer indicating how general CT
/// is.
}
}
+/// Examine constraint type and operand type and determine a weight value,
+/// where: -1 = invalid match, and 0 = so-so match to 3 = good match.
+/// This object must already have been set up with the operand type
+/// and the current alternative constraint selected.
+int TargetLowering::getMultipleConstraintMatchWeight(
+ AsmOperandInfo &info, int maIndex) const {
+ std::vector<std::string> *rCodes;
+ if (maIndex >= (int)info.multipleAlternatives.size())
+ rCodes = &info.Codes;
+ else
+ rCodes = &info.multipleAlternatives[maIndex].Codes;
+ int BestWeight = -1;
+
+ // Loop over the options, keeping track of the most general one.
+ for (unsigned i = 0, e = rCodes->size(); i != e; ++i) {
+ int weight = getSingleConstraintMatchWeight(info, (*rCodes)[i].c_str());
+ if (weight > BestWeight)
+ BestWeight = weight;
+ }
+
+ return BestWeight;
+}
+
+/// Examine constraint type and operand type and determine a weight value,
+/// where: -1 = invalid match, and 0 = so-so match to 3 = good match.
+/// This object must already have been set up with the operand type
+/// and the current alternative constraint selected.
+int TargetLowering::getSingleConstraintMatchWeight(
+ AsmOperandInfo &info, const char *constraint) const {
+ int weight = -1;
+ Value *CallOperandVal = info.CallOperandVal;
+ // If we don't have a value, we can't do a match,
+ // but allow it at the lowest weight.
+ if (CallOperandVal == NULL)
+ return 0;
+ // Look at the constraint type.
+ switch (*constraint) {
+ case 'i': // immediate integer.
+ case 'n': // immediate integer with a known value.
+ weight = 0;
+ if (info.CallOperandVal) {
+ if (isa<ConstantInt>(info.CallOperandVal))
+ weight = 3;
+ else
+ weight = -1;
+ }
+ break;
+ case 's': // non-explicit intregal immediate.
+ weight = 0;
+ if (info.CallOperandVal) {
+ if (isa<GlobalValue>(info.CallOperandVal))
+ weight = 3;
+ else
+ weight = -1;
+ }
+ break;
+ case 'm': // memory operand.
+ case 'o': // offsettable memory operand
+ case 'V': // non-offsettable memory operand
+ weight = 2;
+ break;
+ case 'g': // general register, memory operand or immediate integer.
+ case 'X': // any operand.
+ weight = 1;
+ break;
+ default:
+ weight = 0;
+ break;
+ }
+ return weight;
+}
+
/// ChooseConstraint - If there are multiple different constraints that we
/// could pick for this operand (e.g. "imr") try to pick the 'best' one.
/// This is somewhat tricky: constraints fall into four classes:
/// 'm' over 'r', for example.
///
static void ChooseConstraint(TargetLowering::AsmOperandInfo &OpInfo,
- bool hasMemory, const TargetLowering &TLI,
+ const TargetLowering &TLI,
SDValue Op, SelectionDAG *DAG) {
assert(OpInfo.Codes.size() > 1 && "Doesn't have multiple constraint options");
unsigned BestIdx = 0;
TargetLowering::ConstraintType BestType = TargetLowering::C_Unknown;
int BestGenerality = -1;
-
+
// Loop over the options, keeping track of the most general one.
for (unsigned i = 0, e = OpInfo.Codes.size(); i != e; ++i) {
TargetLowering::ConstraintType CType =
TLI.getConstraintType(OpInfo.Codes[i]);
-
+
// If this is an 'other' constraint, see if the operand is valid for it.
// For example, on X86 we might have an 'rI' constraint. If the operand
// is an integer in the range [0..31] we want to use I (saving a load
assert(OpInfo.Codes[i].size() == 1 &&
"Unhandled multi-letter 'other' constraint");
std::vector<SDValue> ResultOps;
- TLI.LowerAsmOperandForConstraint(Op, OpInfo.Codes[i][0], hasMemory,
+ TLI.LowerAsmOperandForConstraint(Op, OpInfo.Codes[i][0],
ResultOps, *DAG);
if (!ResultOps.empty()) {
BestType = CType;
}
}
+ // Things with matching constraints can only be registers, per gcc
+ // documentation. This mainly affects "g" constraints.
+ if (CType == TargetLowering::C_Memory && OpInfo.hasMatchingInput())
+ continue;
+
// This constraint letter is more general than the previous one, use it.
int Generality = getConstraintGenerality(CType);
if (Generality > BestGenerality) {
/// OpInfo.ConstraintCode and OpInfo.ConstraintType.
void TargetLowering::ComputeConstraintToUse(AsmOperandInfo &OpInfo,
SDValue Op,
- bool hasMemory,
SelectionDAG *DAG) const {
assert(!OpInfo.Codes.empty() && "Must have at least one constraint");
OpInfo.ConstraintCode = OpInfo.Codes[0];
OpInfo.ConstraintType = getConstraintType(OpInfo.ConstraintCode);
} else {
- ChooseConstraint(OpInfo, hasMemory, *this, Op, DAG);
+ ChooseConstraint(OpInfo, *this, Op, DAG);
}
// 'X' matches anything.