#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"
#include "llvm/Support/MathExtras.h"
+#include <cctype>
using namespace llvm;
namespace llvm {
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.
setIndexedLoadAction(IM, (MVT::SimpleValueType)VT, Expand);
setIndexedStoreAction(IM, (MVT::SimpleValueType)VT, Expand);
}
-
+
// These operations default to expand.
setOperationAction(ISD::FGETSIGN, (MVT::SimpleValueType)VT, Expand);
setOperationAction(ISD::CONCAT_VECTORS, (MVT::SimpleValueType)VT, Expand);
// Most targets ignore the @llvm.prefetch intrinsic.
setOperationAction(ISD::PREFETCH, MVT::Other, Expand);
-
- // ConstantFP nodes default to expand. Targets can either change this to
+
+ // ConstantFP nodes default to expand. Targets can either change this to
// Legal, in which case all fp constants are legal, or use isFPImmLegal()
// to optimize expansions for certain constants.
setOperationAction(ISD::ConstantFP, MVT::f32, Expand);
// Default ISD::TRAP to expand (which turns it into abort).
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));
maxStoresPerMemset = maxStoresPerMemcpy = maxStoresPerMemmove = 8;
+ maxStoresPerMemsetOptSize = maxStoresPerMemcpyOptSize
+ = maxStoresPerMemmoveOptSize = 4;
benefitFromCodePlacementOpt = false;
UseUnderscoreSetJmp = false;
UseUnderscoreLongJmp = false;
SelectIsExpensive = false;
IntDivIsCheap = false;
Pow2DivIsCheap = false;
+ JumpIsExpensive = false;
StackPointerRegisterToSaveRestore = 0;
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();
-
+
unsigned NumVectorRegs = 1;
-
- // FIXME: We don't support non-power-of-2-sized vectors for now. Ideally we
+
+ // 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))) {
}
NumIntermediates = NumVectorRegs;
-
+
MVT NewVT = MVT::getVectorVT(EltTy, NumElts);
if (!TLI->isTypeLegal(NewVT))
NewVT = EltTy;
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;
+
+ // 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 1;
+ 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() {
RegisterTypeForVT[MVT::ppcf128] = MVT::f64;
TransformToType[MVT::ppcf128] = MVT::f64;
ValueTypeActions.setTypeAction(MVT::ppcf128, Expand);
- }
+ }
// Decide how to handle f64. If the target does not have native f64 support,
// expand it to i64 and we will be generating soft float library calls.
ValueTypeActions.setTypeAction(MVT::f32, Expand);
}
}
-
+
// Loop over all of the vector value types to see which need transformations.
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) {
+ if (SVT.getVectorElementType() == EltVT &&
+ SVT.getVectorNumElements() > NElts &&
+ isTypeLegal(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;
-
- // FIXME: We don't support non-power-of-2-sized vectors for now. Ideally we
+
+ // 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 && !isTypeLegal(
}
NumIntermediates = NumVectorRegs;
-
+
EVT NewVT = EVT::getVectorVT(Context, EltTy, NumElts);
if (!isTypeLegal(NewVT))
NewVT = EltTy;
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
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.
// Optimization Methods
//===----------------------------------------------------------------------===//
-/// ShrinkDemandedConstant - Check to see if the specified operand of the
+/// ShrinkDemandedConstant - Check to see if the specified operand of the
/// specified instruction is a constant integer. If so, check to see if there
/// are any bits set in the constant that are not demanded. If so, shrink the
/// constant and return true.
-bool TargetLowering::TargetLoweringOpt::ShrinkDemandedConstant(SDValue Op,
+bool TargetLowering::TargetLoweringOpt::ShrinkDemandedConstant(SDValue Op,
const APInt &Demanded) {
DebugLoc dl = Op.getDebugLoc();
EVT VT = Op.getValueType();
SDValue New = DAG.getNode(Op.getOpcode(), dl, VT, Op.getOperand(0),
DAG.getConstant(Demanded &
- C->getAPIntValue(),
+ C->getAPIntValue(),
VT));
return CombineTo(Op, New);
}
KnownZero = KnownOne = APInt(BitWidth, 0);
// Other users may use these bits.
- if (!Op.getNode()->hasOneUse()) {
+ if (!Op.getNode()->hasOneUse()) {
if (Depth != 0) {
- // If not at the root, Just compute the KnownZero/KnownOne bits to
+ // If not at the root, Just compute the KnownZero/KnownOne bits to
// simplify things downstream.
TLO.DAG.ComputeMaskedBits(Op, DemandedMask, KnownZero, KnownOne, Depth);
return false;
// If this is the root being simplified, allow it to have multiple uses,
// just set the NewMask to all bits.
NewMask = APInt::getAllOnesValue(BitWidth);
- } else if (DemandedMask == 0) {
+ } else if (DemandedMask == 0) {
// Not demanding any bits from Op.
if (Op.getOpcode() != ISD::UNDEF)
return TLO.CombineTo(Op, TLO.DAG.getUNDEF(Op.getValueType()));
// the RHS.
if (ConstantSDNode *RHSC = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
APInt LHSZero, LHSOne;
+ // Do not increment Depth here; that can cause an infinite loop.
TLO.DAG.ComputeMaskedBits(Op.getOperand(0), NewMask,
- LHSZero, LHSOne, Depth+1);
+ LHSZero, LHSOne, Depth);
// If the LHS already has zeros where RHSC does, this and is dead.
if ((LHSZero & NewMask) == (~RHSC->getAPIntValue() & NewMask))
return TLO.CombineTo(Op, Op.getOperand(0));
if (TLO.ShrinkDemandedConstant(Op, ~LHSZero & NewMask))
return true;
}
-
+
if (SimplifyDemandedBits(Op.getOperand(1), NewMask, KnownZero,
KnownOne, TLO, Depth+1))
return true;
- assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
+ assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
if (SimplifyDemandedBits(Op.getOperand(0), ~KnownZero & NewMask,
KnownZero2, KnownOne2, TLO, Depth+1))
return true;
- assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
-
+ assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
+
// If all of the demanded bits are known one on one side, return the other.
// These bits cannot contribute to the result of the 'and'.
if ((NewMask & ~KnownZero2 & KnownOne) == (~KnownZero2 & NewMask))
KnownZero |= KnownZero2;
break;
case ISD::OR:
- if (SimplifyDemandedBits(Op.getOperand(1), NewMask, KnownZero,
+ if (SimplifyDemandedBits(Op.getOperand(1), NewMask, KnownZero,
KnownOne, TLO, Depth+1))
return true;
- assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
+ assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
if (SimplifyDemandedBits(Op.getOperand(0), ~KnownOne & NewMask,
KnownZero2, KnownOne2, TLO, Depth+1))
return true;
- assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
-
+ assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
+
// If all of the demanded bits are known zero on one side, return the other.
// These bits cannot contribute to the result of the 'or'.
if ((NewMask & ~KnownOne2 & KnownZero) == (~KnownOne2 & NewMask))
KnownOne |= KnownOne2;
break;
case ISD::XOR:
- if (SimplifyDemandedBits(Op.getOperand(1), NewMask, KnownZero,
+ if (SimplifyDemandedBits(Op.getOperand(1), NewMask, KnownZero,
KnownOne, TLO, Depth+1))
return true;
- assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
+ assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
if (SimplifyDemandedBits(Op.getOperand(0), NewMask, KnownZero2,
KnownOne2, TLO, Depth+1))
return true;
- assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
-
+ assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
+
// If all of the demanded bits are known zero on one side, return the other.
// These bits cannot contribute to the result of the 'xor'.
if ((KnownZero & NewMask) == NewMask)
return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::OR, dl, Op.getValueType(),
Op.getOperand(0),
Op.getOperand(1)));
-
+
// Output known-0 bits are known if clear or set in both the LHS & RHS.
KnownZeroOut = (KnownZero & KnownZero2) | (KnownOne & KnownOne2);
// Output known-1 are known to be set if set in only one of the LHS, RHS.
KnownOneOut = (KnownZero & KnownOne2) | (KnownOne & KnownZero2);
-
+
// If all of the demanded bits on one side are known, and all of the set
// bits on that side are also known to be set on the other side, turn this
// into an AND, as we know the bits will be cleared.
if ((KnownOne & KnownOne2) == KnownOne) {
EVT VT = Op.getValueType();
SDValue ANDC = TLO.DAG.getConstant(~KnownOne & NewMask, VT);
- return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::AND, dl, VT,
+ return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::AND, dl, VT,
Op.getOperand(0), ANDC));
}
}
-
+
// If the RHS is a constant, see if we can simplify it.
// for XOR, we prefer to force bits to 1 if they will make a -1.
// if we can't force bits, try to shrink constant
KnownOne = KnownOneOut;
break;
case ISD::SELECT:
- if (SimplifyDemandedBits(Op.getOperand(2), NewMask, KnownZero,
+ if (SimplifyDemandedBits(Op.getOperand(2), NewMask, KnownZero,
KnownOne, TLO, Depth+1))
return true;
if (SimplifyDemandedBits(Op.getOperand(1), NewMask, KnownZero2,
KnownOne2, TLO, Depth+1))
return true;
- assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
- assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
-
+ assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
+ assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
+
// If the operands are constants, see if we can simplify them.
if (TLO.ShrinkDemandedConstant(Op, NewMask))
return true;
-
+
// Only known if known in both the LHS and RHS.
KnownOne &= KnownOne2;
KnownZero &= KnownZero2;
break;
case ISD::SELECT_CC:
- if (SimplifyDemandedBits(Op.getOperand(3), NewMask, KnownZero,
+ if (SimplifyDemandedBits(Op.getOperand(3), NewMask, KnownZero,
KnownOne, TLO, Depth+1))
return true;
if (SimplifyDemandedBits(Op.getOperand(2), NewMask, KnownZero2,
KnownOne2, TLO, Depth+1))
return true;
- assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
- assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
-
+ assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
+ assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
+
// If the operands are constants, see if we can simplify them.
if (TLO.ShrinkDemandedConstant(Op, NewMask))
return true;
-
+
// Only known if known in both the LHS and RHS.
KnownOne &= KnownOne2;
KnownZero &= KnownZero2;
if (Diff < 0) {
Diff = -Diff;
Opc = ISD::SRL;
- }
-
- SDValue NewSA =
+ }
+
+ SDValue NewSA =
TLO.DAG.getConstant(Diff, Op.getOperand(1).getValueType());
EVT VT = Op.getValueType();
return TLO.CombineTo(Op, TLO.DAG.getNode(Opc, dl, VT,
InOp.getOperand(0), NewSA));
}
- }
-
- 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.
unsigned ShAmt = SA->getZExtValue();
unsigned VTSize = VT.getSizeInBits();
SDValue InOp = Op.getOperand(0);
-
+
// If the shift count is an invalid immediate, don't do anything.
if (ShAmt >= BitWidth)
break;
if (Diff < 0) {
Diff = -Diff;
Opc = ISD::SHL;
- }
-
+ }
+
SDValue NewSA =
TLO.DAG.getConstant(Diff, Op.getOperand(1).getValueType());
return TLO.CombineTo(Op, TLO.DAG.getNode(Opc, dl, VT,
InOp.getOperand(0), NewSA));
}
- }
-
+ }
+
// Compute the new bits that are at the top now.
if (SimplifyDemandedBits(InOp, (NewMask << ShAmt),
KnownZero, KnownOne, TLO, Depth+1))
return true;
- assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
+ assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
KnownZero = KnownZero.lshr(ShAmt);
KnownOne = KnownOne.lshr(ShAmt);
// 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();
unsigned ShAmt = SA->getZExtValue();
-
+
// If the shift count is an invalid immediate, don't do anything.
if (ShAmt >= BitWidth)
break;
APInt HighBits = APInt::getHighBitsSet(BitWidth, ShAmt);
if (HighBits.intersects(NewMask))
InDemandedMask |= APInt::getSignBit(VT.getScalarType().getSizeInBits());
-
+
if (SimplifyDemandedBits(Op.getOperand(0), InDemandedMask,
KnownZero, KnownOne, TLO, Depth+1))
return true;
- assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
+ assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
KnownZero = KnownZero.lshr(ShAmt);
KnownOne = KnownOne.lshr(ShAmt);
-
+
// Handle the sign bit, adjusted to where it is now in the mask.
APInt SignBit = APInt::getSignBit(BitWidth).lshr(ShAmt);
-
+
// If the input sign bit is known to be zero, or if none of the top bits
// are demanded, turn this into an unsigned shift right.
if (KnownZero.intersects(SignBit) || (HighBits & ~NewMask) == HighBits) {
- return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::SRL, dl, VT,
+ return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::SRL, dl, VT,
Op.getOperand(0),
Op.getOperand(1)));
} else if (KnownOne.intersects(SignBit)) { // New bits are known one.
case ISD::SIGN_EXTEND_INREG: {
EVT EVT = cast<VTSDNode>(Op.getOperand(1))->getVT();
- // Sign extension. Compute the demanded bits in the result that are not
+ // 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());
+
// 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.getSizeInBits());
- InSignBit.zext(BitWidth);
- APInt InputDemandedBits = APInt::getLowBitsSet(BitWidth,
- EVT.getSizeInBits()) &
- NewMask;
-
+ APInt InSignBit =
+ APInt::getSignBit(EVT.getScalarType().getSizeInBits()).zext(BitWidth);
+ APInt InputDemandedBits =
+ APInt::getLowBitsSet(BitWidth,
+ EVT.getScalarType().getSizeInBits()) &
+ NewMask;
+
// Since the sign extended bits are demanded, we know that the sign
// bit is demanded.
InputDemandedBits |= InSignBit;
if (SimplifyDemandedBits(Op.getOperand(0), InputDemandedBits,
KnownZero, KnownOne, TLO, Depth+1))
return true;
- assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
+ assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
// If the sign bit of the input is known set or clear, then we know the
// top bits of the result.
-
+
// If the input sign bit is known zero, convert this into a zero extension.
if (KnownZero.intersects(InSignBit))
- return TLO.CombineTo(Op,
+ return TLO.CombineTo(Op,
TLO.DAG.getZeroExtendInReg(Op.getOperand(0),dl,EVT));
-
+
if (KnownOne.intersects(InSignBit)) { // Input sign bit known set
KnownOne |= NewBits;
KnownZero &= ~NewBits;
break;
}
case ISD::ZERO_EXTEND: {
- unsigned OperandBitWidth = Op.getOperand(0).getValueSizeInBits();
- APInt InMask = NewMask;
- InMask.trunc(OperandBitWidth);
-
+ unsigned OperandBitWidth =
+ Op.getOperand(0).getValueType().getScalarType().getSizeInBits();
+ APInt InMask = NewMask.trunc(OperandBitWidth);
+
// If none of the top bits are demanded, convert this into an any_extend.
APInt NewBits =
APInt::getHighBitsSet(BitWidth, BitWidth - OperandBitWidth) & NewMask;
if (!NewBits.intersects(NewMask))
return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::ANY_EXTEND, dl,
- Op.getValueType(),
+ Op.getValueType(),
Op.getOperand(0)));
-
+
if (SimplifyDemandedBits(Op.getOperand(0), InMask,
KnownZero, KnownOne, TLO, Depth+1))
return true;
- assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
- KnownZero.zext(BitWidth);
- KnownOne.zext(BitWidth);
+ assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
+ KnownZero = KnownZero.zext(BitWidth);
+ KnownOne = KnownOne.zext(BitWidth);
KnownZero |= NewBits;
break;
}
case ISD::SIGN_EXTEND: {
EVT InVT = Op.getOperand(0).getValueType();
- unsigned InBits = InVT.getSizeInBits();
+ unsigned InBits = InVT.getScalarType().getSizeInBits();
APInt InMask = APInt::getLowBitsSet(BitWidth, InBits);
APInt InSignBit = APInt::getBitsSet(BitWidth, InBits - 1, InBits);
APInt NewBits = ~InMask & NewMask;
-
+
// If none of the top bits are demanded, convert this into an any_extend.
if (NewBits == 0)
return TLO.CombineTo(Op,TLO.DAG.getNode(ISD::ANY_EXTEND, dl,
Op.getValueType(),
Op.getOperand(0)));
-
+
// Since some of the sign extended bits are demanded, we know that the sign
// bit is demanded.
APInt InDemandedBits = InMask & NewMask;
InDemandedBits |= InSignBit;
- InDemandedBits.trunc(InBits);
-
- if (SimplifyDemandedBits(Op.getOperand(0), InDemandedBits, KnownZero,
+ InDemandedBits = InDemandedBits.trunc(InBits);
+
+ if (SimplifyDemandedBits(Op.getOperand(0), InDemandedBits, KnownZero,
KnownOne, TLO, Depth+1))
return true;
- KnownZero.zext(BitWidth);
- KnownOne.zext(BitWidth);
-
+ KnownZero = KnownZero.zext(BitWidth);
+ KnownOne = KnownOne.zext(BitWidth);
+
// If the sign bit is known zero, convert this to a zero extend.
if (KnownZero.intersects(InSignBit))
return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::ZERO_EXTEND, dl,
- Op.getValueType(),
+ Op.getValueType(),
Op.getOperand(0)));
-
+
// If the sign bit is known one, the top bits match.
if (KnownOne.intersects(InSignBit)) {
KnownOne |= NewBits;
break;
}
case ISD::ANY_EXTEND: {
- unsigned OperandBitWidth = Op.getOperand(0).getValueSizeInBits();
- APInt InMask = NewMask;
- InMask.trunc(OperandBitWidth);
+ unsigned OperandBitWidth =
+ Op.getOperand(0).getValueType().getScalarType().getSizeInBits();
+ APInt InMask = NewMask.trunc(OperandBitWidth);
if (SimplifyDemandedBits(Op.getOperand(0), InMask,
KnownZero, KnownOne, TLO, Depth+1))
return true;
- assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
- KnownZero.zext(BitWidth);
- KnownOne.zext(BitWidth);
+ assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
+ KnownZero = KnownZero.zext(BitWidth);
+ KnownOne = KnownOne.zext(BitWidth);
break;
}
case ISD::TRUNCATE: {
// Simplify the input, using demanded bit information, and compute the known
// zero/one bits live out.
- APInt TruncMask = NewMask;
- TruncMask.zext(Op.getOperand(0).getValueSizeInBits());
+ unsigned OperandBitWidth =
+ Op.getOperand(0).getValueType().getScalarType().getSizeInBits();
+ APInt TruncMask = NewMask.zext(OperandBitWidth);
if (SimplifyDemandedBits(Op.getOperand(0), TruncMask,
KnownZero, KnownOne, TLO, Depth+1))
return true;
- KnownZero.trunc(BitWidth);
- KnownOne.trunc(BitWidth);
-
+ KnownZero = KnownZero.trunc(BitWidth);
+ KnownOne = KnownOne.trunc(BitWidth);
+
// If the input is only used by this truncate, see if we can shrink it based
// 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()).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;
}
}
-
- assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
+
+ assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
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,
+ 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());
- 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?");
KnownZero |= ~InMask & NewMask;
break;
}
- case ISD::BIT_CONVERT:
+ case ISD::BITCAST:
#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.
isOperationLegal(ISD::FGETSIGN, Op.getValueType())) {
// 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, Op.getValueType(),
+ SDValue Sign = TLO.DAG.getNode(ISD::FGETSIGN, Op.getValueType(),
Op.getOperand(0));
unsigned ShVal = Op.getValueType().getSizeInBits()-1;
SDValue ShAmt = TLO.DAG.getConstant(ShVal, getShiftAmountTy());
TLO.DAG.ComputeMaskedBits(Op, NewMask, KnownZero, KnownOne, Depth);
break;
}
-
+
// If we know the value of all of the demanded bits, return this as a
// constant.
if ((NewMask & (KnownZero|KnownOne)) == NewMask)
return TLO.CombineTo(Op, TLO.DAG.getConstant(KnownOne, Op.getValueType()));
-
+
return false;
}
-/// computeMaskedBitsForTargetNode - Determine which of the bits specified
-/// in Mask are known to be either zero or one and return them in the
+/// computeMaskedBitsForTargetNode - Determine which of the bits specified
+/// in Mask are known to be either zero or one and return them in the
/// KnownZero/KnownOne bitsets.
-void TargetLowering::computeMaskedBitsForTargetNode(const SDValue Op,
+void TargetLowering::computeMaskedBitsForTargetNode(const SDValue Op,
const APInt &Mask,
- APInt &KnownZero,
+ APInt &KnownZero,
APInt &KnownOne,
const SelectionDAG &DAG,
unsigned Depth) const {
// 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);
(KnownOne.countPopulation() == 1);
}
-/// SimplifySetCC - Try to simplify a setcc built with the specified operands
+/// 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(EVT VT, SDValue N0, SDValue N1,
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)) {
}
}
+ SDValue CTPOP = N0;
+ // Look through truncs that don't change the value of a ctpop.
+ if (N0.hasOneUse() && N0.getOpcode() == ISD::TRUNCATE)
+ CTPOP = N0.getOperand(0);
+
+ if (CTPOP.hasOneUse() && CTPOP.getOpcode() == ISD::CTPOP &&
+ (N0 == CTPOP || N0.getValueType().getSizeInBits() >
+ Log2_32_Ceil(CTPOP.getValueType().getSizeInBits()))) {
+ EVT CTVT = CTPOP.getValueType();
+ SDValue CTOp = CTPOP.getOperand(0);
+
+ // (ctpop x) u< 2 -> (x & x-1) == 0
+ // (ctpop x) u> 1 -> (x & x-1) != 0
+ if ((Cond == ISD::SETULT && C1 == 2) || (Cond == ISD::SETUGT && C1 == 1)){
+ SDValue Sub = DAG.getNode(ISD::SUB, dl, CTVT, CTOp,
+ DAG.getConstant(1, CTVT));
+ SDValue And = DAG.getNode(ISD::AND, dl, CTVT, CTOp, Sub);
+ ISD::CondCode CC = Cond == ISD::SETULT ? ISD::SETEQ : ISD::SETNE;
+ return DAG.getSetCC(dl, VT, And, DAG.getConstant(0, CTVT), CC);
+ }
+
+ // TODO: (ctpop x) == 1 -> x && (x & x-1) == 0 iff ctpop is illegal.
+ }
+
// If the LHS is '(and load, const)', the RHS is 0,
// the test is for equality or unsigned, and all 1 bits of the const are
// in the same partial word, see if we can shorten the load.
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();
- // We can narrow (e.g.) 16-bit extending loads on 32-bit target to
+ 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;
}
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);
- return DAG.getSetCC(dl, VT,
+ Lod->getPointerInfo().getWithOffset(bestOffset),
+ 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);
}
}
(isOperationLegal(ISD::SETCC, newVT) &&
getCondCodeAction(Cond, newVT)==Legal))
return DAG.getSetCC(dl, VT, N0.getOperand(0),
- DAG.getConstant(APInt(C1).trunc(InSize), newVT),
+ DAG.getConstant(C1.trunc(InSize), newVT),
Cond);
break;
}
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;
EVT Op0Ty = N0.getOperand(0).getValueType();
if (Op0Ty == ExtSrcTy) {
if (!DCI.isCalledByLegalizer())
DCI.AddToWorklist(ZextOp.getNode());
// Otherwise, make this a use of a zext.
- return DAG.getSetCC(dl, VT, ZextOp,
+ return DAG.getSetCC(dl, VT, ZextOp,
DAG.getConstant(C1 & APInt::getLowBitsSet(
ExtDstTyBits,
- ExtSrcTyBits),
+ ExtSrcTyBits),
ExtDstTy),
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,
+ 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)) &&
if (N0.getOpcode() == ISD::XOR)
Val = N0.getOperand(0);
else {
- assert(N0.getOpcode() == ISD::AND &&
+ assert(N0.getOpcode() == ISD::AND &&
N0.getOperand(0).getOpcode() == ISD::XOR);
// ((X^1)&1)^1 -> X & 1
Val = DAG.getNode(ISD::AND, dl, N0.getValueType(),
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);
+ }
}
}
-
+
APInt MinVal, MaxVal;
unsigned OperandBitSize = N1C->getValueType(0).getSizeInBits();
if (ISD::isSignedIntSetCC(Cond)) {
if (Cond == ISD::SETGE || Cond == ISD::SETUGE) {
if (C1 == MinVal) return DAG.getConstant(1, VT); // X >= MIN --> true
// X >= C0 --> X > (C0-1)
- return DAG.getSetCC(dl, VT, N0,
+ return DAG.getSetCC(dl, VT, N0,
DAG.getConstant(C1-1, N1.getValueType()),
(Cond == ISD::SETGE) ? ISD::SETGT : ISD::SETUGT);
}
if (Cond == ISD::SETLE || Cond == ISD::SETULE) {
if (C1 == MaxVal) return DAG.getConstant(1, VT); // X <= MAX --> true
// X <= C0 --> X < (C0+1)
- return DAG.getSetCC(dl, VT, N0,
+ return DAG.getSetCC(dl, VT, N0,
DAG.getConstant(C1+1, N1.getValueType()),
(Cond == ISD::SETLE) ? ISD::SETLT : ISD::SETULT);
}
// If we have setult X, 1, turn it into seteq X, 0
if ((Cond == ISD::SETLT || Cond == ISD::SETULT) && C1 == MinVal+1)
- return DAG.getSetCC(dl, VT, N0,
- DAG.getConstant(MinVal, N0.getValueType()),
+ return DAG.getSetCC(dl, VT, N0,
+ DAG.getConstant(MinVal, N0.getValueType()),
ISD::SETEQ);
// If we have setugt X, Max-1, turn it into seteq X, Max
else if ((Cond == ISD::SETGT || Cond == ISD::SETUGT) && C1 == MaxVal-1)
- return DAG.getSetCC(dl, VT, N0,
+ return DAG.getSetCC(dl, VT, N0,
DAG.getConstant(MaxVal, N0.getValueType()),
ISD::SETEQ);
// by changing cc.
// SETUGT X, SINTMAX -> SETLT X, 0
- if (Cond == ISD::SETUGT &&
+ if (Cond == ISD::SETUGT &&
C1 == APInt::getSignedMaxValue(OperandBitSize))
- return DAG.getSetCC(dl, VT, N0,
+ return DAG.getSetCC(dl, VT, N0,
DAG.getConstant(0, N1.getValueType()),
ISD::SETLT);
// 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))) {
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)));
}
}
}
return DAG.getUNDEF(VT);
}
}
-
+
// Otherwise, we know the RHS is not a NaN. Simplify the node to drop the
// constant if knowing that the operand is non-nan is enough. We prefer to
// have SETO(x,x) instead of SETO(x, 0.0) because this avoids having to
if (DAG.isCommutativeBinOp(N0.getOpcode())) {
// If X op Y == Y op X, try other combinations.
if (N0.getOperand(0) == N1.getOperand(1))
- return DAG.getSetCC(dl, VT, N0.getOperand(1), N1.getOperand(0),
+ return DAG.getSetCC(dl, VT, N0.getOperand(1), N1.getOperand(0),
Cond);
if (N0.getOperand(1) == N1.getOperand(0))
- return DAG.getSetCC(dl, VT, N0.getOperand(0), N1.getOperand(1),
+ return DAG.getSetCC(dl, VT, N0.getOperand(0), N1.getOperand(1),
Cond);
}
}
-
+
if (ConstantSDNode *RHSC = dyn_cast<ConstantSDNode>(N1)) {
if (ConstantSDNode *LHSR = dyn_cast<ConstantSDNode>(N0.getOperand(1))) {
// Turn (X+C1) == C2 --> X == C2-C1
LHSR->getAPIntValue(),
N0.getValueType()), Cond);
}
-
+
// Turn (X^C1) == C2 into X == C1^C2 iff X&~C1 = 0.
if (N0.getOpcode() == ISD::XOR)
// If we know that all of the inverted bits are zero, don't bother
N0.getValueType()),
Cond);
}
-
+
// Turn (C1-X) == C2 --> X == C1-C2
if (ConstantSDNode *SUBC = dyn_cast<ConstantSDNode>(N0.getOperand(0))) {
if (N0.getOpcode() == ISD::SUB && N0.getNode()->hasOneUse()) {
N0.getValueType()),
Cond);
}
- }
+ }
}
// Simplify (X+Z) == X --> Z == 0
assert(N0.getOpcode() == ISD::SUB && "Unexpected operation!");
// (Z-X) == X --> Z == X<<1
SDValue SH = DAG.getNode(ISD::SHL, dl, N1.getValueType(),
- N1,
+ N1,
DAG.getConstant(1, getShiftAmountTy()));
if (!DCI.isCalledByLegalizer())
DCI.AddToWorklist(SH.getNode());
} else if (N1.getNode()->hasOneUse()) {
assert(N1.getOpcode() == ISD::SUB && "Unexpected operation!");
// X == (Z-X) --> X<<1 == Z
- SDValue SH = DAG.getNode(ISD::SHL, dl, N1.getValueType(), N0,
+ SDValue SH = DAG.getNode(ISD::SHL, dl, N1.getValueType(), N0,
DAG.getConstant(1, getShiftAmountTy()));
if (!DCI.isCalledByLegalizer())
DCI.AddToWorklist(SH.getNode());
/// 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;
}
}
-
- if (Constraint.size() > 1 && Constraint[0] == '{' &&
+
+ if (Constraint.size() > 1 && Constraint[0] == '{' &&
Constraint[Constraint.size()-1] == '}')
return C_Register;
return C_Unknown;
/// 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) {
// is possible and fine if either GV or C are missing.
ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op);
GlobalAddressSDNode *GA = dyn_cast<GlobalAddressSDNode>(Op);
-
+
// If we have "(add GV, C)", pull out GV/C
if (Op.getOpcode() == ISD::ADD) {
C = dyn_cast<ConstantSDNode>(Op.getOperand(1));
if (C == 0 || GA == 0)
C = 0, GA = 0;
}
-
+
// If we find a valid operand, map to the TargetXXX version so that the
// value itself doesn't get selected.
if (GA) { // Either &GV or &GV+C
int64_t Offs = GA->getOffset();
if (C) Offs += C->getZExtValue();
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.
for (TargetRegisterInfo::regclass_iterator RCI = RI->regclass_begin(),
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();
break;
}
}
-
+
if (!isLegal) continue;
-
- for (TargetRegisterClass::iterator I = RC->begin(), E = RC->end();
+
+ for (TargetRegisterClass::iterator I = RC->begin(), E = RC->end();
I != E; ++I) {
if (RegName.equals_lower(RI->getName(*I)))
return std::make_pair(*I, RC);
}
}
-
- return std::pair<unsigned, const TargetRegisterClass*>(0, 0);
+
+ return std::make_pair(0u, static_cast<const TargetRegisterClass*>(0));
}
//===----------------------------------------------------------------------===//
}
+/// 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.
+TargetLowering::AsmOperandInfoVector TargetLowering::ParseConstraints(
+ ImmutableCallSite CS) const {
+ /// ConstraintOperands - Information about all of the constraints.
+ AsmOperandInfoVector 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.
+ InlineAsm::ConstraintInfoVector
+ 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();
+
+ OpInfo.ConstraintVT = 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())) {
+ OpInfo.ConstraintVT = getValueType(STy->getElementType(ResNo));
+ } else {
+ assert(ResNo == 0 && "Asm only has one result!");
+ OpInfo.ConstraintVT = 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 (OpInfo.CallOperandVal) {
+ const llvm::Type *OpTy = OpInfo.CallOperandVal->getType();
+ if (OpInfo.isIndirect) {
+ const llvm::PointerType *PtrTy = dyn_cast<PointerType>(OpTy);
+ if (!PtrTy)
+ report_fatal_error("Indirect operand for inline asm not a pointer!");
+ OpTy = PtrTy->getElementType();
+ }
+ // If OpTy is not a single value, it may be a struct/union that we
+ // can tile with integers.
+ if (!OpTy->isSingleValueType() && OpTy->isSized()) {
+ unsigned BitSize = TD->getTypeSizeInBits(OpTy);
+ switch (BitSize) {
+ default: break;
+ case 1:
+ case 8:
+ case 16:
+ case 32:
+ case 64:
+ case 128:
+ OpInfo.ConstraintVT =
+ EVT::getEVT(IntegerType::get(OpTy->getContext(), BitSize), true);
+ break;
+ }
+ } else if (dyn_cast<PointerType>(OpTy)) {
+ OpInfo.ConstraintVT = MVT::getIntegerVT(8*TD->getPointerSize());
+ } else {
+ OpInfo.ConstraintVT = EVT::getEVT(OpTy, true);
+ }
+ }
+ }
+
+ // 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;
+ }
+ }
+ }
+ 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!");
+ }
+ }
+
+ }
+ }
+
+ return ConstraintOperands;
+}
+
+
/// getConstraintGenerality - Return an integer indicating how general CT
/// is.
static unsigned getConstraintGenerality(TargetLowering::ConstraintType CT) {
}
}
+/// Examine constraint type and operand type and determine a weight value.
+/// This object must already have been set up with the operand type
+/// and the current alternative constraint selected.
+TargetLowering::ConstraintWeight
+ TargetLowering::getMultipleConstraintMatchWeight(
+ AsmOperandInfo &info, int maIndex) const {
+ InlineAsm::ConstraintCodeVector *rCodes;
+ if (maIndex >= (int)info.multipleAlternatives.size())
+ rCodes = &info.Codes;
+ else
+ rCodes = &info.multipleAlternatives[maIndex].Codes;
+ ConstraintWeight BestWeight = CW_Invalid;
+
+ // Loop over the options, keeping track of the most general one.
+ for (unsigned i = 0, e = rCodes->size(); i != e; ++i) {
+ ConstraintWeight 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.
+/// This object must already have been set up with the operand type
+/// and the current alternative constraint selected.
+TargetLowering::ConstraintWeight
+ TargetLowering::getSingleConstraintMatchWeight(
+ AsmOperandInfo &info, const char *constraint) const {
+ ConstraintWeight weight = CW_Invalid;
+ 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 CW_Default;
+ // Look at the constraint type.
+ switch (*constraint) {
+ case 'i': // immediate integer.
+ case 'n': // immediate integer with a known value.
+ if (isa<ConstantInt>(CallOperandVal))
+ weight = CW_Constant;
+ break;
+ case 's': // non-explicit intregal immediate.
+ if (isa<GlobalValue>(CallOperandVal))
+ weight = CW_Constant;
+ break;
+ case 'E': // immediate float if host format.
+ case 'F': // immediate float.
+ if (isa<ConstantFP>(CallOperandVal))
+ weight = CW_Constant;
+ break;
+ case '<': // memory operand with autodecrement.
+ case '>': // memory operand with autoincrement.
+ case 'm': // memory operand.
+ case 'o': // offsettable memory operand
+ case 'V': // non-offsettable memory operand
+ weight = CW_Memory;
+ break;
+ case 'r': // general register.
+ case 'g': // general register, memory operand or immediate integer.
+ // note: Clang converts "g" to "imr".
+ if (CallOperandVal->getType()->isIntegerTy())
+ weight = CW_Register;
+ break;
+ case 'X': // any operand.
+ default:
+ weight = CW_Default;
+ 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;
break;
}
}
-
+
+ // 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) {
BestGenerality = Generality;
}
}
-
+
OpInfo.ConstraintCode = OpInfo.Codes[BestIdx];
OpInfo.ConstraintType = BestType;
}
/// type to use for the specific AsmOperandInfo, setting
/// OpInfo.ConstraintCode and OpInfo.ConstraintType.
void TargetLowering::ComputeConstraintToUse(AsmOperandInfo &OpInfo,
- SDValue Op,
- bool hasMemory,
+ SDValue Op,
SelectionDAG *DAG) const {
assert(!OpInfo.Codes.empty() && "Must have at least one constraint");
-
+
// Single-letter constraints ('r') are very common.
if (OpInfo.Codes.size() == 1) {
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.
if (OpInfo.ConstraintCode == "X" && OpInfo.CallOperandVal) {
// Labels and constants are handled elsewhere ('X' is the only thing
OpInfo.CallOperandVal = v;
return;
}
-
+
// Otherwise, try to resolve it to something we know about by looking at
// the actual operand type.
if (const char *Repl = LowerXConstraint(OpInfo.ConstraintVT)) {
/// 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,
+bool TargetLowering::isLegalAddressingMode(const AddrMode &AM,
const Type *Ty) const {
// The default implementation of this implements a conservative RISCy, r+r and
// r+i addr mode.
// Allows a sign-extended 16-bit immediate field.
if (AM.BaseOffs <= -(1LL << 16) || AM.BaseOffs >= (1LL << 16)-1)
return false;
-
+
// No global is ever allowed as a base.
if (AM.BaseGV)
return false;
-
- // Only support r+r,
+
+ // Only support r+r,
switch (AM.Scale) {
case 0: // "r+i" or just "i", depending on HasBaseReg.
break;
// Allow 2*r as r+r.
break;
}
-
+
return true;
}
/// 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,
+SDValue TargetLowering::BuildSDIV(SDNode *N, SelectionDAG &DAG,
std::vector<SDNode*>* Created) const {
EVT VT = N->getValueType(0);
DebugLoc dl= N->getDebugLoc();
-
+
// Check to see if we can do this.
// FIXME: We should be more aggressive here.
if (!isTypeLegal(VT))
return SDValue();
-
+
APInt d = cast<ConstantSDNode>(N->getOperand(1))->getAPIntValue();
APInt::ms magics = d.magic();
-
+
// Multiply the numerator (operand 0) by the magic value
// FIXME: We should support doing a MUL in a wider type
SDValue Q;
else
return SDValue(); // No mulhs or equvialent
// If d > 0 and m < 0, add the numerator
- if (d.isStrictlyPositive() && magics.m.isNegative()) {
+ if (d.isStrictlyPositive() && magics.m.isNegative()) {
Q = DAG.getNode(ISD::ADD, dl, VT, Q, N->getOperand(0));
if (Created)
Created->push_back(Q.getNode());
}
// Shift right algebraic if shift value is nonzero
if (magics.s > 0) {
- Q = DAG.getNode(ISD::SRA, dl, VT, Q,
+ Q = DAG.getNode(ISD::SRA, dl, VT, Q,
DAG.getConstant(magics.s, getShiftAmountTy()));
if (Created)
Created->push_back(Q.getNode());
if (magics.a == 0) {
assert(magics.s < N1C->getAPIntValue().getBitWidth() &&
"We shouldn't generate an undefined shift!");
- return DAG.getNode(ISD::SRL, dl, VT, Q,
+ return DAG.getNode(ISD::SRL, dl, VT, Q,
DAG.getConstant(magics.s, getShiftAmountTy()));
} else {
SDValue NPQ = DAG.getNode(ISD::SUB, dl, VT, N->getOperand(0), Q);
if (Created)
Created->push_back(NPQ.getNode());
- NPQ = DAG.getNode(ISD::SRL, dl, VT, NPQ,
+ NPQ = DAG.getNode(ISD::SRL, dl, VT, NPQ,
DAG.getConstant(1, getShiftAmountTy()));
if (Created)
Created->push_back(NPQ.getNode());
NPQ = DAG.getNode(ISD::ADD, dl, VT, NPQ, Q);
if (Created)
Created->push_back(NPQ.getNode());
- return DAG.getNode(ISD::SRL, dl, VT, NPQ,
+ return DAG.getNode(ISD::SRL, dl, VT, NPQ,
DAG.getConstant(magics.s-1, getShiftAmountTy()));
}
}