#include "ARMConstantPoolValue.h"
#include "ARMISelLowering.h"
#include "ARMMachineFunctionInfo.h"
+#include "ARMPerfectShuffle.h"
#include "ARMRegisterInfo.h"
#include "ARMSubtarget.h"
#include "ARMTargetMachine.h"
#include "llvm/ADT/VectorExtras.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MathExtras.h"
+#include <sstream>
using namespace llvm;
-static bool CC_ARM_APCS_Custom_f64(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
+static bool CC_ARM_APCS_Custom_f64(unsigned &ValNo, EVT &ValVT, EVT &LocVT,
CCValAssign::LocInfo &LocInfo,
ISD::ArgFlagsTy &ArgFlags,
CCState &State);
-static bool CC_ARM_AAPCS_Custom_f64(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
+static bool CC_ARM_AAPCS_Custom_f64(unsigned &ValNo, EVT &ValVT, EVT &LocVT,
CCValAssign::LocInfo &LocInfo,
ISD::ArgFlagsTy &ArgFlags,
CCState &State);
-static bool RetCC_ARM_APCS_Custom_f64(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
+static bool RetCC_ARM_APCS_Custom_f64(unsigned &ValNo, EVT &ValVT, EVT &LocVT,
CCValAssign::LocInfo &LocInfo,
ISD::ArgFlagsTy &ArgFlags,
CCState &State);
-static bool RetCC_ARM_AAPCS_Custom_f64(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
+static bool RetCC_ARM_AAPCS_Custom_f64(unsigned &ValNo, EVT &ValVT, EVT &LocVT,
CCValAssign::LocInfo &LocInfo,
ISD::ArgFlagsTy &ArgFlags,
CCState &State);
-void ARMTargetLowering::addTypeForNEON(MVT VT, MVT PromotedLdStVT,
- MVT PromotedBitwiseVT) {
+void ARMTargetLowering::addTypeForNEON(EVT VT, EVT PromotedLdStVT,
+ EVT PromotedBitwiseVT) {
if (VT != PromotedLdStVT) {
- setOperationAction(ISD::LOAD, VT, Promote);
- AddPromotedToType (ISD::LOAD, VT, PromotedLdStVT);
+ setOperationAction(ISD::LOAD, VT.getSimpleVT(), Promote);
+ AddPromotedToType (ISD::LOAD, VT.getSimpleVT(),
+ PromotedLdStVT.getSimpleVT());
- setOperationAction(ISD::STORE, VT, Promote);
- AddPromotedToType (ISD::STORE, VT, PromotedLdStVT);
+ setOperationAction(ISD::STORE, VT.getSimpleVT(), Promote);
+ AddPromotedToType (ISD::STORE, VT.getSimpleVT(),
+ PromotedLdStVT.getSimpleVT());
}
- MVT ElemTy = VT.getVectorElementType();
+ EVT ElemTy = VT.getVectorElementType();
if (ElemTy != MVT::i64 && ElemTy != MVT::f64)
- setOperationAction(ISD::VSETCC, VT, Custom);
+ setOperationAction(ISD::VSETCC, VT.getSimpleVT(), Custom);
if (ElemTy == MVT::i8 || ElemTy == MVT::i16)
- setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Custom);
- setOperationAction(ISD::BUILD_VECTOR, VT, Custom);
- setOperationAction(ISD::VECTOR_SHUFFLE, VT, Custom);
- setOperationAction(ISD::SCALAR_TO_VECTOR, VT, Custom);
- setOperationAction(ISD::CONCAT_VECTORS, VT, Custom);
+ setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT.getSimpleVT(), Custom);
+ if (ElemTy != MVT::i32) {
+ setOperationAction(ISD::SINT_TO_FP, VT.getSimpleVT(), Expand);
+ setOperationAction(ISD::UINT_TO_FP, VT.getSimpleVT(), Expand);
+ setOperationAction(ISD::FP_TO_SINT, VT.getSimpleVT(), Expand);
+ setOperationAction(ISD::FP_TO_UINT, VT.getSimpleVT(), Expand);
+ }
+ setOperationAction(ISD::BUILD_VECTOR, VT.getSimpleVT(), Custom);
+ setOperationAction(ISD::VECTOR_SHUFFLE, VT.getSimpleVT(), Custom);
+ setOperationAction(ISD::CONCAT_VECTORS, VT.getSimpleVT(), Custom);
+ setOperationAction(ISD::EXTRACT_SUBVECTOR, VT.getSimpleVT(), Expand);
if (VT.isInteger()) {
- setOperationAction(ISD::SHL, VT, Custom);
- setOperationAction(ISD::SRA, VT, Custom);
- setOperationAction(ISD::SRL, VT, Custom);
+ setOperationAction(ISD::SHL, VT.getSimpleVT(), Custom);
+ setOperationAction(ISD::SRA, VT.getSimpleVT(), Custom);
+ setOperationAction(ISD::SRL, VT.getSimpleVT(), Custom);
}
// Promote all bit-wise operations.
if (VT.isInteger() && VT != PromotedBitwiseVT) {
- setOperationAction(ISD::AND, VT, Promote);
- AddPromotedToType (ISD::AND, VT, PromotedBitwiseVT);
- setOperationAction(ISD::OR, VT, Promote);
- AddPromotedToType (ISD::OR, VT, PromotedBitwiseVT);
- setOperationAction(ISD::XOR, VT, Promote);
- AddPromotedToType (ISD::XOR, VT, PromotedBitwiseVT);
- }
-}
-
-void ARMTargetLowering::addDRTypeForNEON(MVT VT) {
+ setOperationAction(ISD::AND, VT.getSimpleVT(), Promote);
+ AddPromotedToType (ISD::AND, VT.getSimpleVT(),
+ PromotedBitwiseVT.getSimpleVT());
+ setOperationAction(ISD::OR, VT.getSimpleVT(), Promote);
+ AddPromotedToType (ISD::OR, VT.getSimpleVT(),
+ PromotedBitwiseVT.getSimpleVT());
+ setOperationAction(ISD::XOR, VT.getSimpleVT(), Promote);
+ AddPromotedToType (ISD::XOR, VT.getSimpleVT(),
+ PromotedBitwiseVT.getSimpleVT());
+ }
+
+ // Neon does not support vector divide/remainder operations.
+ setOperationAction(ISD::SDIV, VT.getSimpleVT(), Expand);
+ setOperationAction(ISD::UDIV, VT.getSimpleVT(), Expand);
+ setOperationAction(ISD::FDIV, VT.getSimpleVT(), Expand);
+ setOperationAction(ISD::SREM, VT.getSimpleVT(), Expand);
+ setOperationAction(ISD::UREM, VT.getSimpleVT(), Expand);
+ setOperationAction(ISD::FREM, VT.getSimpleVT(), Expand);
+}
+
+void ARMTargetLowering::addDRTypeForNEON(EVT VT) {
addRegisterClass(VT, ARM::DPRRegisterClass);
addTypeForNEON(VT, MVT::f64, MVT::v2i32);
}
-void ARMTargetLowering::addQRTypeForNEON(MVT VT) {
+void ARMTargetLowering::addQRTypeForNEON(EVT VT) {
addRegisterClass(VT, ARM::QPRRegisterClass);
addTypeForNEON(VT, MVT::v2f64, MVT::v4i32);
}
setLibcallName(RTLIB::SRL_I128, 0);
setLibcallName(RTLIB::SRA_I128, 0);
+ // Libcalls should use the AAPCS base standard ABI, even if hard float
+ // is in effect, as per the ARM RTABI specification, section 4.1.2.
+ if (Subtarget->isAAPCS_ABI()) {
+ for (int i = 0; i < RTLIB::UNKNOWN_LIBCALL; ++i) {
+ setLibcallCallingConv(static_cast<RTLIB::Libcall>(i),
+ CallingConv::ARM_AAPCS);
+ }
+ }
+
if (Subtarget->isThumb1Only())
addRegisterClass(MVT::i32, ARM::tGPRRegisterClass);
else
addQRTypeForNEON(MVT::v4i32);
addQRTypeForNEON(MVT::v2i64);
+ // v2f64 is legal so that QR subregs can be extracted as f64 elements, but
+ // neither Neon nor VFP support any arithmetic operations on it.
+ setOperationAction(ISD::FADD, MVT::v2f64, Expand);
+ setOperationAction(ISD::FSUB, MVT::v2f64, Expand);
+ setOperationAction(ISD::FMUL, MVT::v2f64, Expand);
+ setOperationAction(ISD::FDIV, MVT::v2f64, Expand);
+ setOperationAction(ISD::FREM, MVT::v2f64, Expand);
+ setOperationAction(ISD::FCOPYSIGN, MVT::v2f64, Expand);
+ setOperationAction(ISD::VSETCC, MVT::v2f64, Expand);
+ setOperationAction(ISD::FNEG, MVT::v2f64, Expand);
+ setOperationAction(ISD::FABS, MVT::v2f64, Expand);
+ setOperationAction(ISD::FSQRT, MVT::v2f64, Expand);
+ setOperationAction(ISD::FSIN, MVT::v2f64, Expand);
+ setOperationAction(ISD::FCOS, MVT::v2f64, Expand);
+ setOperationAction(ISD::FPOWI, MVT::v2f64, Expand);
+ setOperationAction(ISD::FPOW, MVT::v2f64, Expand);
+ setOperationAction(ISD::FLOG, MVT::v2f64, Expand);
+ setOperationAction(ISD::FLOG2, MVT::v2f64, Expand);
+ setOperationAction(ISD::FLOG10, MVT::v2f64, Expand);
+ setOperationAction(ISD::FEXP, MVT::v2f64, Expand);
+ setOperationAction(ISD::FEXP2, MVT::v2f64, Expand);
+ setOperationAction(ISD::FCEIL, MVT::v2f64, Expand);
+ setOperationAction(ISD::FTRUNC, MVT::v2f64, Expand);
+ setOperationAction(ISD::FRINT, MVT::v2f64, Expand);
+ setOperationAction(ISD::FNEARBYINT, MVT::v2f64, Expand);
+ setOperationAction(ISD::FFLOOR, MVT::v2f64, Expand);
+
+ // Neon does not support some operations on v1i64 and v2i64 types.
+ setOperationAction(ISD::MUL, MVT::v1i64, Expand);
+ setOperationAction(ISD::MUL, MVT::v2i64, Expand);
+ setOperationAction(ISD::VSETCC, MVT::v1i64, Expand);
+ setOperationAction(ISD::VSETCC, MVT::v2i64, Expand);
+
setTargetDAGCombine(ISD::INTRINSIC_WO_CHAIN);
setTargetDAGCombine(ISD::SHL);
setTargetDAGCombine(ISD::SRL);
setOperationAction(ISD::VAEND, MVT::Other, Expand);
setOperationAction(ISD::STACKSAVE, MVT::Other, Expand);
setOperationAction(ISD::STACKRESTORE, MVT::Other, Expand);
+ setOperationAction(ISD::EHSELECTION, MVT::i32, Expand);
+ // FIXME: Shouldn't need this, since no register is used, but the legalizer
+ // doesn't yet know how to not do that for SjLj.
+ setExceptionSelectorRegister(ARM::R0);
if (Subtarget->isThumb())
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Custom);
else
// We want to custom lower some of our intrinsics.
setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);
- setOperationAction(ISD::INTRINSIC_W_CHAIN, MVT::Other, Custom);
- setOperationAction(ISD::INTRINSIC_VOID, MVT::Other, Custom);
setOperationAction(ISD::SETCC, MVT::i32, Expand);
setOperationAction(ISD::SETCC, MVT::f32, Expand);
setStackPointerRegisterToSaveRestore(ARM::SP);
setSchedulingPreference(SchedulingForRegPressure);
- setIfCvtBlockSizeLimit(Subtarget->isThumb() ? 0 : 10);
- setIfCvtDupBlockSizeLimit(Subtarget->isThumb() ? 0 : 2);
-
- if (!Subtarget->isThumb()) {
- // Use branch latency information to determine if-conversion limits.
- // FIXME: If-converter should use instruction latency of the branch being
- // eliminated to compute the threshold. For ARMv6, the branch "latency"
- // varies depending on whether it's dynamically or statically predicted
- // and on whether the destination is in the prefetch buffer.
- const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
- const InstrItineraryData &InstrItins = Subtarget->getInstrItineraryData();
- unsigned Latency= InstrItins.getLatency(TII->get(ARM::Bcc).getSchedClass());
- if (Latency > 1) {
- setIfCvtBlockSizeLimit(Latency-1);
- if (Latency > 2)
- setIfCvtDupBlockSizeLimit(Latency-2);
- } else {
- setIfCvtBlockSizeLimit(10);
- setIfCvtDupBlockSizeLimit(2);
- }
+
+ // FIXME: If-converter should use instruction latency to determine
+ // profitability rather than relying on fixed limits.
+ if (Subtarget->getCPUString() == "generic") {
+ // Generic (and overly aggressive) if-conversion limits.
+ setIfCvtBlockSizeLimit(10);
+ setIfCvtDupBlockSizeLimit(2);
+ } else if (Subtarget->hasV6Ops()) {
+ setIfCvtBlockSizeLimit(2);
+ setIfCvtDupBlockSizeLimit(1);
+ } else {
+ setIfCvtBlockSizeLimit(3);
+ setIfCvtDupBlockSizeLimit(2);
}
maxStoresPerMemcpy = 1; //// temporary - rewrite interface to use type
case ARMISD::VQRSHRNsu: return "ARMISD::VQRSHRNsu";
case ARMISD::VGETLANEu: return "ARMISD::VGETLANEu";
case ARMISD::VGETLANEs: return "ARMISD::VGETLANEs";
- case ARMISD::VDUPLANEQ: return "ARMISD::VDUPLANEQ";
- case ARMISD::VLD2D: return "ARMISD::VLD2D";
- case ARMISD::VLD3D: return "ARMISD::VLD3D";
- case ARMISD::VLD4D: return "ARMISD::VLD4D";
- case ARMISD::VST2D: return "ARMISD::VST2D";
- case ARMISD::VST3D: return "ARMISD::VST3D";
- case ARMISD::VST4D: return "ARMISD::VST4D";
+ case ARMISD::VDUP: return "ARMISD::VDUP";
+ case ARMISD::VDUPLANE: return "ARMISD::VDUPLANE";
+ case ARMISD::VEXT: return "ARMISD::VEXT";
+ case ARMISD::VREV64: return "ARMISD::VREV64";
+ case ARMISD::VREV32: return "ARMISD::VREV32";
+ case ARMISD::VREV16: return "ARMISD::VREV16";
+ case ARMISD::VZIP: return "ARMISD::VZIP";
+ case ARMISD::VUZP: return "ARMISD::VUZP";
+ case ARMISD::VTRN: return "ARMISD::VTRN";
}
}
/// getFunctionAlignment - Return the Log2 alignment of this function.
unsigned ARMTargetLowering::getFunctionAlignment(const Function *F) const {
- return getTargetMachine().getSubtarget<ARMSubtarget>().isThumb() ? 1 : 2;
+ return getTargetMachine().getSubtarget<ARMSubtarget>().isThumb() ? 0 : 1;
}
//===----------------------------------------------------------------------===//
}
}
-/// FPCCToARMCC - Convert a DAG fp condition code to an ARM CC. It
-/// returns true if the operands should be inverted to form the proper
-/// comparison.
-static bool FPCCToARMCC(ISD::CondCode CC, ARMCC::CondCodes &CondCode,
+/// FPCCToARMCC - Convert a DAG fp condition code to an ARM CC.
+static void FPCCToARMCC(ISD::CondCode CC, ARMCC::CondCodes &CondCode,
ARMCC::CondCodes &CondCode2) {
- bool Invert = false;
CondCode2 = ARMCC::AL;
switch (CC) {
default: llvm_unreachable("Unknown FP condition!");
case ISD::SETGE:
case ISD::SETOGE: CondCode = ARMCC::GE; break;
case ISD::SETOLT: CondCode = ARMCC::MI; break;
- case ISD::SETOLE: CondCode = ARMCC::GT; Invert = true; break;
+ case ISD::SETOLE: CondCode = ARMCC::LS; break;
case ISD::SETONE: CondCode = ARMCC::MI; CondCode2 = ARMCC::GT; break;
case ISD::SETO: CondCode = ARMCC::VC; break;
case ISD::SETUO: CondCode = ARMCC::VS; break;
case ISD::SETNE:
case ISD::SETUNE: CondCode = ARMCC::NE; break;
}
- return Invert;
}
//===----------------------------------------------------------------------===//
#include "ARMGenCallingConv.inc"
// APCS f64 is in register pairs, possibly split to stack
-static bool f64AssignAPCS(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
+static bool f64AssignAPCS(unsigned &ValNo, EVT &ValVT, EVT &LocVT,
CCValAssign::LocInfo &LocInfo,
CCState &State, bool CanFail) {
static const unsigned RegList[] = { ARM::R0, ARM::R1, ARM::R2, ARM::R3 };
return true;
}
-static bool CC_ARM_APCS_Custom_f64(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
+static bool CC_ARM_APCS_Custom_f64(unsigned &ValNo, EVT &ValVT, EVT &LocVT,
CCValAssign::LocInfo &LocInfo,
ISD::ArgFlagsTy &ArgFlags,
CCState &State) {
}
// AAPCS f64 is in aligned register pairs
-static bool f64AssignAAPCS(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
+static bool f64AssignAAPCS(unsigned &ValNo, EVT &ValVT, EVT &LocVT,
CCValAssign::LocInfo &LocInfo,
CCState &State, bool CanFail) {
static const unsigned HiRegList[] = { ARM::R0, ARM::R2 };
return true;
}
-static bool CC_ARM_AAPCS_Custom_f64(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
+static bool CC_ARM_AAPCS_Custom_f64(unsigned &ValNo, EVT &ValVT, EVT &LocVT,
CCValAssign::LocInfo &LocInfo,
ISD::ArgFlagsTy &ArgFlags,
CCState &State) {
return true; // we handled it
}
-static bool f64RetAssign(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
+static bool f64RetAssign(unsigned &ValNo, EVT &ValVT, EVT &LocVT,
CCValAssign::LocInfo &LocInfo, CCState &State) {
static const unsigned HiRegList[] = { ARM::R0, ARM::R2 };
static const unsigned LoRegList[] = { ARM::R1, ARM::R3 };
return true;
}
-static bool RetCC_ARM_APCS_Custom_f64(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
+static bool RetCC_ARM_APCS_Custom_f64(unsigned &ValNo, EVT &ValVT, EVT &LocVT,
CCValAssign::LocInfo &LocInfo,
ISD::ArgFlagsTy &ArgFlags,
CCState &State) {
return true; // we handled it
}
-static bool RetCC_ARM_AAPCS_Custom_f64(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
+static bool RetCC_ARM_AAPCS_Custom_f64(unsigned &ValNo, EVT &ValVT, EVT &LocVT,
CCValAssign::LocInfo &LocInfo,
ISD::ArgFlagsTy &ArgFlags,
CCState &State) {
/// CCAssignFnForNode - Selects the correct CCAssignFn for a the
/// given CallingConvention value.
-CCAssignFn *ARMTargetLowering::CCAssignFnForNode(unsigned CC,
+CCAssignFn *ARMTargetLowering::CCAssignFnForNode(CallingConv::ID CC,
bool Return,
bool isVarArg) const {
switch (CC) {
/// appropriate copies out of appropriate physical registers.
SDValue
ARMTargetLowering::LowerCallResult(SDValue Chain, SDValue InFlag,
- unsigned CallConv, bool isVarArg,
+ CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins,
DebugLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) {
/// nodes.
SDValue
ARMTargetLowering::LowerCall(SDValue Chain, SDValue Callee,
- unsigned CallConv, bool isVarArg,
+ CallingConv::ID CallConv, bool isVarArg,
bool isTailCall,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<ISD::InputArg> &Ins,
isLocalARMFunc = !Subtarget->isThumb() && !isExt;
// tBX takes a register source operand.
if (isARMFunc && Subtarget->isThumb1Only() && !Subtarget->hasV5TOps()) {
- ARMConstantPoolValue *CPV = new ARMConstantPoolValue(GV, ARMPCLabelIndex,
- ARMCP::CPStub, 4);
+ ARMConstantPoolValue *CPV = new ARMConstantPoolValue(GV,
+ ARMPCLabelIndex,
+ ARMCP::CPValue, 4);
SDValue CPAddr = DAG.getTargetConstantPool(CPV, getPointerTy(), 4);
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
Callee = DAG.getLoad(getPointerTy(), dl,
// tBX takes a register source operand.
const char *Sym = S->getSymbol();
if (isARMFunc && Subtarget->isThumb1Only() && !Subtarget->hasV5TOps()) {
- ARMConstantPoolValue *CPV = new ARMConstantPoolValue(Sym, ARMPCLabelIndex,
- ARMCP::CPStub, 4);
+ ARMConstantPoolValue *CPV = new ARMConstantPoolValue(*DAG.getContext(),
+ Sym, ARMPCLabelIndex, 4);
SDValue CPAddr = DAG.getTargetConstantPool(CPV, getPointerTy(), 4);
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
Callee = DAG.getLoad(getPointerTy(), dl,
SDValue
ARMTargetLowering::LowerReturn(SDValue Chain,
- unsigned CallConv, bool isVarArg,
+ CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::OutputArg> &Outs,
DebugLoc dl, SelectionDAG &DAG) {
// be used to form addressing mode. These wrapped nodes will be selected
// into MOVi.
static SDValue LowerConstantPool(SDValue Op, SelectionDAG &DAG) {
- MVT PtrVT = Op.getValueType();
+ EVT PtrVT = Op.getValueType();
// FIXME there is no actual debug info here
DebugLoc dl = Op.getDebugLoc();
ConstantPoolSDNode *CP = cast<ConstantPoolSDNode>(Op);
ARMTargetLowering::LowerToTLSGeneralDynamicModel(GlobalAddressSDNode *GA,
SelectionDAG &DAG) {
DebugLoc dl = GA->getDebugLoc();
- MVT PtrVT = getPointerTy();
+ EVT PtrVT = getPointerTy();
unsigned char PCAdj = Subtarget->isThumb() ? 4 : 8;
ARMConstantPoolValue *CPV =
- new ARMConstantPoolValue(GA->getGlobal(), ARMPCLabelIndex, ARMCP::CPValue,
- PCAdj, "tlsgd", true);
+ new ARMConstantPoolValue(GA->getGlobal(), ARMPCLabelIndex,
+ ARMCP::CPValue, PCAdj, "tlsgd", true);
SDValue Argument = DAG.getTargetConstantPool(CPV, PtrVT, 4);
Argument = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, Argument);
Argument = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), Argument, NULL, 0);
ArgListTy Args;
ArgListEntry Entry;
Entry.Node = Argument;
- Entry.Ty = (const Type *) Type::Int32Ty;
+ Entry.Ty = (const Type *) Type::getInt32Ty(*DAG.getContext());
Args.push_back(Entry);
// FIXME: is there useful debug info available here?
std::pair<SDValue, SDValue> CallResult =
- LowerCallTo(Chain, (const Type *) Type::Int32Ty, false, false, false, false,
+ LowerCallTo(Chain, (const Type *) Type::getInt32Ty(*DAG.getContext()),
+ false, false, false, false,
0, CallingConv::C, false, /*isReturnValueUsed=*/true,
DAG.getExternalSymbol("__tls_get_addr", PtrVT), Args, DAG, dl);
return CallResult.first;
DebugLoc dl = GA->getDebugLoc();
SDValue Offset;
SDValue Chain = DAG.getEntryNode();
- MVT PtrVT = getPointerTy();
+ EVT PtrVT = getPointerTy();
// Get the Thread Pointer
SDValue ThreadPointer = DAG.getNode(ARMISD::THREAD_POINTER, dl, PtrVT);
// initial exec model
unsigned char PCAdj = Subtarget->isThumb() ? 4 : 8;
ARMConstantPoolValue *CPV =
- new ARMConstantPoolValue(GA->getGlobal(), ARMPCLabelIndex, ARMCP::CPValue,
- PCAdj, "gottpoff", true);
+ new ARMConstantPoolValue(GA->getGlobal(), ARMPCLabelIndex,
+ ARMCP::CPValue, PCAdj, "gottpoff", true);
Offset = DAG.getTargetConstantPool(CPV, PtrVT, 4);
Offset = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, Offset);
Offset = DAG.getLoad(PtrVT, dl, Chain, Offset, NULL, 0);
Offset = DAG.getLoad(PtrVT, dl, Chain, Offset, NULL, 0);
} else {
// local exec model
- ARMConstantPoolValue *CPV =
- new ARMConstantPoolValue(GV, ARMCP::CPValue, "tpoff");
+ ARMConstantPoolValue *CPV = new ARMConstantPoolValue(GV, "tpoff");
Offset = DAG.getTargetConstantPool(CPV, PtrVT, 4);
Offset = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, Offset);
Offset = DAG.getLoad(PtrVT, dl, Chain, Offset, NULL, 0);
SDValue ARMTargetLowering::LowerGlobalAddressELF(SDValue Op,
SelectionDAG &DAG) {
- MVT PtrVT = getPointerTy();
+ EVT PtrVT = getPointerTy();
DebugLoc dl = Op.getDebugLoc();
GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
Reloc::Model RelocM = getTargetMachine().getRelocationModel();
if (RelocM == Reloc::PIC_) {
bool UseGOTOFF = GV->hasLocalLinkage() || GV->hasHiddenVisibility();
ARMConstantPoolValue *CPV =
- new ARMConstantPoolValue(GV, ARMCP::CPValue, UseGOTOFF ? "GOTOFF":"GOT");
+ new ARMConstantPoolValue(GV, UseGOTOFF ? "GOTOFF" : "GOT");
SDValue CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4);
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
SDValue Result = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(),
- CPAddr, NULL, 0);
+ CPAddr,
+ PseudoSourceValue::getConstantPool(), 0);
SDValue Chain = Result.getValue(1);
SDValue GOT = DAG.getGLOBAL_OFFSET_TABLE(PtrVT);
Result = DAG.getNode(ISD::ADD, dl, PtrVT, Result, GOT);
if (!UseGOTOFF)
- Result = DAG.getLoad(PtrVT, dl, Chain, Result, NULL, 0);
+ Result = DAG.getLoad(PtrVT, dl, Chain, Result,
+ PseudoSourceValue::getGOT(), 0);
return Result;
} else {
SDValue CPAddr = DAG.getTargetConstantPool(GV, PtrVT, 4);
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
- return DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), CPAddr, NULL, 0);
+ return DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), CPAddr,
+ PseudoSourceValue::getConstantPool(), 0);
}
}
-/// GVIsIndirectSymbol - true if the GV will be accessed via an indirect symbol
-/// even in non-static mode.
-static bool GVIsIndirectSymbol(GlobalValue *GV, Reloc::Model RelocM) {
- // If symbol visibility is hidden, the extra load is not needed if
- // the symbol is definitely defined in the current translation unit.
- bool isDecl = GV->isDeclaration() || GV->hasAvailableExternallyLinkage();
- if (GV->hasHiddenVisibility() && (!isDecl && !GV->hasCommonLinkage()))
- return false;
- return RelocM != Reloc::Static && (isDecl || GV->isWeakForLinker());
-}
-
SDValue ARMTargetLowering::LowerGlobalAddressDarwin(SDValue Op,
SelectionDAG &DAG) {
- MVT PtrVT = getPointerTy();
+ EVT PtrVT = getPointerTy();
DebugLoc dl = Op.getDebugLoc();
GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
Reloc::Model RelocM = getTargetMachine().getRelocationModel();
- bool IsIndirect = GVIsIndirectSymbol(GV, RelocM);
SDValue CPAddr;
if (RelocM == Reloc::Static)
CPAddr = DAG.getTargetConstantPool(GV, PtrVT, 4);
else {
- unsigned PCAdj = (RelocM != Reloc::PIC_)
- ? 0 : (Subtarget->isThumb() ? 4 : 8);
- ARMCP::ARMCPKind Kind = IsIndirect ? ARMCP::CPNonLazyPtr
- : ARMCP::CPValue;
- ARMConstantPoolValue *CPV = new ARMConstantPoolValue(GV, ARMPCLabelIndex,
- Kind, PCAdj);
+ unsigned PCAdj = (RelocM != Reloc::PIC_) ? 0 : (Subtarget->isThumb()?4:8);
+ ARMConstantPoolValue *CPV =
+ new ARMConstantPoolValue(GV, ARMPCLabelIndex, ARMCP::CPValue, PCAdj);
CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4);
}
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex++, MVT::i32);
Result = DAG.getNode(ARMISD::PIC_ADD, dl, PtrVT, Result, PICLabel);
}
- if (IsIndirect)
+
+ if (Subtarget->GVIsIndirectSymbol(GV, RelocM))
Result = DAG.getLoad(PtrVT, dl, Chain, Result, NULL, 0);
return Result;
SelectionDAG &DAG){
assert(Subtarget->isTargetELF() &&
"GLOBAL OFFSET TABLE not implemented for non-ELF targets");
- MVT PtrVT = getPointerTy();
+ EVT PtrVT = getPointerTy();
DebugLoc dl = Op.getDebugLoc();
unsigned PCAdj = Subtarget->isThumb() ? 4 : 8;
- ARMConstantPoolValue *CPV = new ARMConstantPoolValue("_GLOBAL_OFFSET_TABLE_",
- ARMPCLabelIndex,
- ARMCP::CPValue, PCAdj);
+ ARMConstantPoolValue *CPV = new ARMConstantPoolValue(*DAG.getContext(),
+ "_GLOBAL_OFFSET_TABLE_",
+ ARMPCLabelIndex, PCAdj);
SDValue CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4);
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
- SDValue Result = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), CPAddr, NULL, 0);
+ SDValue Result = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), CPAddr,
+ PseudoSourceValue::getConstantPool(), 0);
SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex++, MVT::i32);
return DAG.getNode(ARMISD::PIC_ADD, dl, PtrVT, Result, PICLabel);
}
-static SDValue LowerNeonVLDIntrinsic(SDValue Op, SelectionDAG &DAG,
- unsigned Opcode) {
- SDNode *Node = Op.getNode();
- MVT VT = Node->getValueType(0);
- DebugLoc dl = Op.getDebugLoc();
-
- if (!VT.is64BitVector())
- return SDValue(); // unimplemented
-
- SDValue Ops[] = { Node->getOperand(0),
- Node->getOperand(2) };
- return DAG.getNode(Opcode, dl, Node->getVTList(), Ops, 2);
-}
-
-static SDValue LowerNeonVSTIntrinsic(SDValue Op, SelectionDAG &DAG,
- unsigned Opcode, unsigned NumVecs) {
- SDNode *Node = Op.getNode();
- MVT VT = Node->getOperand(3).getValueType();
- DebugLoc dl = Op.getDebugLoc();
-
- if (!VT.is64BitVector())
- return SDValue(); // unimplemented
-
- SmallVector<SDValue, 6> Ops;
- Ops.push_back(Node->getOperand(0));
- Ops.push_back(Node->getOperand(2));
- for (unsigned N = 0; N < NumVecs; ++N)
- Ops.push_back(Node->getOperand(N + 3));
- return DAG.getNode(Opcode, dl, MVT::Other, Ops.data(), Ops.size());
-}
-
-SDValue
-ARMTargetLowering::LowerINTRINSIC_W_CHAIN(SDValue Op, SelectionDAG &DAG) {
- unsigned IntNo = cast<ConstantSDNode>(Op.getOperand(1))->getZExtValue();
- switch (IntNo) {
- case Intrinsic::arm_neon_vld2i:
- case Intrinsic::arm_neon_vld2f:
- return LowerNeonVLDIntrinsic(Op, DAG, ARMISD::VLD2D);
- case Intrinsic::arm_neon_vld3i:
- case Intrinsic::arm_neon_vld3f:
- return LowerNeonVLDIntrinsic(Op, DAG, ARMISD::VLD3D);
- case Intrinsic::arm_neon_vld4i:
- case Intrinsic::arm_neon_vld4f:
- return LowerNeonVLDIntrinsic(Op, DAG, ARMISD::VLD4D);
- case Intrinsic::arm_neon_vst2i:
- case Intrinsic::arm_neon_vst2f:
- return LowerNeonVSTIntrinsic(Op, DAG, ARMISD::VST2D, 2);
- case Intrinsic::arm_neon_vst3i:
- case Intrinsic::arm_neon_vst3f:
- return LowerNeonVSTIntrinsic(Op, DAG, ARMISD::VST3D, 3);
- case Intrinsic::arm_neon_vst4i:
- case Intrinsic::arm_neon_vst4f:
- return LowerNeonVSTIntrinsic(Op, DAG, ARMISD::VST4D, 4);
- default: return SDValue(); // Don't custom lower most intrinsics.
- }
-}
-
SDValue
ARMTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) {
unsigned IntNo = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
switch (IntNo) {
default: return SDValue(); // Don't custom lower most intrinsics.
case Intrinsic::arm_thread_pointer: {
- MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
return DAG.getNode(ARMISD::THREAD_POINTER, dl, PtrVT);
}
+ case Intrinsic::eh_sjlj_lsda: {
+ MachineFunction &MF = DAG.getMachineFunction();
+ EVT PtrVT = getPointerTy();
+ DebugLoc dl = Op.getDebugLoc();
+ Reloc::Model RelocM = getTargetMachine().getRelocationModel();
+ SDValue CPAddr;
+ unsigned PCAdj = (RelocM != Reloc::PIC_)
+ ? 0 : (Subtarget->isThumb() ? 4 : 8);
+ ARMConstantPoolValue *CPV =
+ new ARMConstantPoolValue(MF.getFunction(), ARMPCLabelIndex,
+ ARMCP::CPLSDA, PCAdj);
+ CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4);
+ CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
+ SDValue Result =
+ DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), CPAddr, NULL, 0);
+ SDValue Chain = Result.getValue(1);
+
+ if (RelocM == Reloc::PIC_) {
+ SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex++, MVT::i32);
+ Result = DAG.getNode(ARMISD::PIC_ADD, dl, PtrVT, Result, PICLabel);
+ }
+ return Result;
+ }
case Intrinsic::eh_sjlj_setjmp:
return DAG.getNode(ARMISD::EH_SJLJ_SETJMP, dl, MVT::i32, Op.getOperand(1));
}
// vastart just stores the address of the VarArgsFrameIndex slot into the
// memory location argument.
DebugLoc dl = Op.getDebugLoc();
- MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
SDValue FR = DAG.getFrameIndex(VarArgsFrameIndex, PtrVT);
const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
return DAG.getStore(Op.getOperand(0), dl, FR, Op.getOperand(1), SV, 0);
ARMTargetLowering::LowerDYNAMIC_STACKALLOC(SDValue Op, SelectionDAG &DAG) {
SDNode *Node = Op.getNode();
DebugLoc dl = Node->getDebugLoc();
- MVT VT = Node->getValueType(0);
+ EVT VT = Node->getValueType(0);
SDValue Chain = Op.getOperand(0);
SDValue Size = Op.getOperand(1);
SDValue Align = Op.getOperand(2);
SDValue
ARMTargetLowering::LowerFormalArguments(SDValue Chain,
- unsigned CallConv, bool isVarArg,
+ CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg>
&Ins,
DebugLoc dl, SelectionDAG &DAG,
// Arguments stored in registers.
if (VA.isRegLoc()) {
- MVT RegVT = VA.getLocVT();
+ EVT RegVT = VA.getLocVT();
SDValue ArgValue;
if (VA.needsCustom()) {
static SDValue LowerSELECT_CC(SDValue Op, SelectionDAG &DAG,
const ARMSubtarget *ST) {
- MVT VT = Op.getValueType();
+ EVT VT = Op.getValueType();
SDValue LHS = Op.getOperand(0);
SDValue RHS = Op.getOperand(1);
ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(4))->get();
}
ARMCC::CondCodes CondCode, CondCode2;
- if (FPCCToARMCC(CC, CondCode, CondCode2))
- std::swap(TrueVal, FalseVal);
+ FPCCToARMCC(CC, CondCode, CondCode2);
SDValue ARMCC = DAG.getConstant(CondCode, MVT::i32);
SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
assert(LHS.getValueType() == MVT::f32 || LHS.getValueType() == MVT::f64);
ARMCC::CondCodes CondCode, CondCode2;
- if (FPCCToARMCC(CC, CondCode, CondCode2))
- // Swap the LHS/RHS of the comparison if needed.
- std::swap(LHS, RHS);
+ FPCCToARMCC(CC, CondCode, CondCode2);
SDValue Cmp = getVFPCmp(LHS, RHS, DAG, dl);
SDValue ARMCC = DAG.getConstant(CondCode, MVT::i32);
SDValue Index = Op.getOperand(2);
DebugLoc dl = Op.getDebugLoc();
- MVT PTy = getPointerTy();
+ EVT PTy = getPointerTy();
JumpTableSDNode *JT = cast<JumpTableSDNode>(Table);
ARMFunctionInfo *AFI = DAG.getMachineFunction().getInfo<ARMFunctionInfo>();
SDValue UId = DAG.getConstant(AFI->createJumpTableUId(), PTy);
Addr, Op.getOperand(2), JTI, UId);
}
if (getTargetMachine().getRelocationModel() == Reloc::PIC_) {
- Addr = DAG.getLoad((MVT)MVT::i32, dl, Chain, Addr, NULL, 0);
+ Addr = DAG.getLoad((EVT)MVT::i32, dl, Chain, Addr, NULL, 0);
Chain = Addr.getValue(1);
Addr = DAG.getNode(ISD::ADD, dl, PTy, Addr, Table);
return DAG.getNode(ARMISD::BR_JT, dl, MVT::Other, Chain, Addr, JTI, UId);
}
static SDValue LowerINT_TO_FP(SDValue Op, SelectionDAG &DAG) {
- MVT VT = Op.getValueType();
+ EVT VT = Op.getValueType();
DebugLoc dl = Op.getDebugLoc();
unsigned Opc =
Op.getOpcode() == ISD::SINT_TO_FP ? ARMISD::SITOF : ARMISD::UITOF;
SDValue Tmp0 = Op.getOperand(0);
SDValue Tmp1 = Op.getOperand(1);
DebugLoc dl = Op.getDebugLoc();
- MVT VT = Op.getValueType();
- MVT SrcVT = Tmp1.getValueType();
+ EVT VT = Op.getValueType();
+ EVT SrcVT = Tmp1.getValueType();
SDValue AbsVal = DAG.getNode(ISD::FABS, dl, VT, Tmp0);
SDValue Cmp = getVFPCmp(Tmp1, DAG.getConstantFP(0.0, SrcVT), DAG, dl);
SDValue ARMCC = DAG.getConstant(ARMCC::LT, MVT::i32);
SDValue ARMTargetLowering::LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) {
MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
MFI->setFrameAddressIsTaken(true);
- MVT VT = Op.getValueType();
+ EVT VT = Op.getValueType();
DebugLoc dl = Op.getDebugLoc(); // FIXME probably not meaningful
unsigned Depth = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
unsigned FrameReg = (Subtarget->isThumb() || Subtarget->isTargetDarwin())
unsigned BytesLeft = SizeVal & 3;
unsigned NumMemOps = SizeVal >> 2;
unsigned EmittedNumMemOps = 0;
- MVT VT = MVT::i32;
+ EVT VT = MVT::i32;
unsigned VTSize = 4;
unsigned i = 0;
const unsigned MAX_LOADS_IN_LDM = 6;
/// getZeroVector - Returns a vector of specified type with all zero elements.
///
-static SDValue getZeroVector(MVT VT, SelectionDAG &DAG, DebugLoc dl) {
+static SDValue getZeroVector(EVT VT, SelectionDAG &DAG, DebugLoc dl) {
assert(VT.isVector() && "Expected a vector type");
// Zero vectors are used to represent vector negation and in those cases
// will be implemented with the NEON VNEG instruction. However, VNEG does
// not support i64 elements, so sometimes the zero vectors will need to be
// explicitly constructed. For those cases, and potentially other uses in
- // the future, always build zero vectors as <4 x i32> or <2 x i32> bitcasted
+ // the future, always build zero vectors as <16 x i8> or <8 x i8> bitcasted
// to their dest type. This ensures they get CSE'd.
SDValue Vec;
- SDValue Cst = DAG.getTargetConstant(0, MVT::i32);
- if (VT.getSizeInBits() == 64)
- Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v2i32, Cst, Cst);
- else
- Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, Cst, Cst, Cst, Cst);
+ SDValue Cst = DAG.getTargetConstant(0, MVT::i8);
+ SmallVector<SDValue, 8> Ops;
+ MVT TVT;
+
+ if (VT.getSizeInBits() == 64) {
+ Ops.assign(8, Cst); TVT = MVT::v8i8;
+ } else {
+ Ops.assign(16, Cst); TVT = MVT::v16i8;
+ }
+ Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, TVT, &Ops[0], Ops.size());
return DAG.getNode(ISD::BIT_CONVERT, dl, VT, Vec);
}
/// getOnesVector - Returns a vector of specified type with all bits set.
///
-static SDValue getOnesVector(MVT VT, SelectionDAG &DAG, DebugLoc dl) {
+static SDValue getOnesVector(EVT VT, SelectionDAG &DAG, DebugLoc dl) {
assert(VT.isVector() && "Expected a vector type");
- // Always build ones vectors as <4 x i32> or <2 x i32> bitcasted to their dest
- // type. This ensures they get CSE'd.
+ // Always build ones vectors as <16 x i32> or <8 x i32> bitcasted to their
+ // dest type. This ensures they get CSE'd.
SDValue Vec;
- SDValue Cst = DAG.getTargetConstant(~0U, MVT::i32);
- if (VT.getSizeInBits() == 64)
- Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v2i32, Cst, Cst);
- else
- Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, Cst, Cst, Cst, Cst);
+ SDValue Cst = DAG.getTargetConstant(0xFF, MVT::i8);
+ SmallVector<SDValue, 8> Ops;
+ MVT TVT;
+
+ if (VT.getSizeInBits() == 64) {
+ Ops.assign(8, Cst); TVT = MVT::v8i8;
+ } else {
+ Ops.assign(16, Cst); TVT = MVT::v16i8;
+ }
+ Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, TVT, &Ops[0], Ops.size());
return DAG.getNode(ISD::BIT_CONVERT, dl, VT, Vec);
}
static SDValue LowerShift(SDNode *N, SelectionDAG &DAG,
const ARMSubtarget *ST) {
- MVT VT = N->getValueType(0);
+ EVT VT = N->getValueType(0);
DebugLoc dl = N->getDebugLoc();
// Lower vector shifts on NEON to use VSHL.
// NEON uses the same intrinsics for both left and right shifts. For
// right shifts, the shift amounts are negative, so negate the vector of
// shift amounts.
- MVT ShiftVT = N->getOperand(1).getValueType();
+ EVT ShiftVT = N->getOperand(1).getValueType();
SDValue NegatedCount = DAG.getNode(ISD::SUB, dl, ShiftVT,
getZeroVector(ShiftVT, DAG, dl),
N->getOperand(1));
N->getOperand(0), NegatedCount);
}
- assert(VT == MVT::i64 &&
- (N->getOpcode() == ISD::SRL || N->getOpcode() == ISD::SRA) &&
+ // We can get here for a node like i32 = ISD::SHL i32, i64
+ if (VT != MVT::i64)
+ return SDValue();
+
+ assert((N->getOpcode() == ISD::SRL || N->getOpcode() == ISD::SRA) &&
"Unknown shift to lower!");
// We only lower SRA, SRL of 1 here, all others use generic lowering.
SDValue Op0 = Op.getOperand(0);
SDValue Op1 = Op.getOperand(1);
SDValue CC = Op.getOperand(2);
- MVT VT = Op.getValueType();
+ EVT VT = Op.getValueType();
ISD::CondCode SetCCOpcode = cast<CondCodeSDNode>(CC)->get();
DebugLoc dl = Op.getDebugLoc();
SplatBitSize, DAG);
}
+static bool isVEXTMask(const SmallVectorImpl<int> &M, EVT VT,
+ bool &ReverseVEXT, unsigned &Imm) {
+ unsigned NumElts = VT.getVectorNumElements();
+ ReverseVEXT = false;
+ Imm = M[0];
+
+ // If this is a VEXT shuffle, the immediate value is the index of the first
+ // element. The other shuffle indices must be the successive elements after
+ // the first one.
+ unsigned ExpectedElt = Imm;
+ for (unsigned i = 1; i < NumElts; ++i) {
+ // Increment the expected index. If it wraps around, it may still be
+ // a VEXT but the source vectors must be swapped.
+ ExpectedElt += 1;
+ if (ExpectedElt == NumElts * 2) {
+ ExpectedElt = 0;
+ ReverseVEXT = true;
+ }
+
+ if (ExpectedElt != static_cast<unsigned>(M[i]))
+ return false;
+ }
+
+ // Adjust the index value if the source operands will be swapped.
+ if (ReverseVEXT)
+ Imm -= NumElts;
+
+ return true;
+}
+
/// isVREVMask - Check if a vector shuffle corresponds to a VREV
/// instruction with the specified blocksize. (The order of the elements
/// within each block of the vector is reversed.)
-bool ARM::isVREVMask(ShuffleVectorSDNode *N, unsigned BlockSize) {
+static bool isVREVMask(const SmallVectorImpl<int> &M, EVT VT,
+ unsigned BlockSize) {
assert((BlockSize==16 || BlockSize==32 || BlockSize==64) &&
"Only possible block sizes for VREV are: 16, 32, 64");
- MVT VT = N->getValueType(0);
unsigned NumElts = VT.getVectorNumElements();
unsigned EltSz = VT.getVectorElementType().getSizeInBits();
- unsigned BlockElts = N->getMaskElt(0) + 1;
+ unsigned BlockElts = M[0] + 1;
if (BlockSize <= EltSz || BlockSize != BlockElts * EltSz)
return false;
for (unsigned i = 0; i < NumElts; ++i) {
- if ((unsigned) N->getMaskElt(i) !=
+ if ((unsigned) M[i] !=
(i - i%BlockElts) + (BlockElts - 1 - i%BlockElts))
return false;
}
return true;
}
-static SDValue BuildSplat(SDValue Val, MVT VT, SelectionDAG &DAG, DebugLoc dl) {
+static bool isVTRNMask(const SmallVectorImpl<int> &M, EVT VT,
+ unsigned &WhichResult) {
+ unsigned NumElts = VT.getVectorNumElements();
+ WhichResult = (M[0] == 0 ? 0 : 1);
+ for (unsigned i = 0; i < NumElts; i += 2) {
+ if ((unsigned) M[i] != i + WhichResult ||
+ (unsigned) M[i+1] != i + NumElts + WhichResult)
+ return false;
+ }
+ return true;
+}
+
+static bool isVUZPMask(const SmallVectorImpl<int> &M, EVT VT,
+ unsigned &WhichResult) {
+ unsigned NumElts = VT.getVectorNumElements();
+ WhichResult = (M[0] == 0 ? 0 : 1);
+ for (unsigned i = 0; i != NumElts; ++i) {
+ if ((unsigned) M[i] != 2 * i + WhichResult)
+ return false;
+ }
+
+ // VUZP.32 for 64-bit vectors is a pseudo-instruction alias for VTRN.32.
+ if (VT.is64BitVector() && VT.getVectorElementType().getSizeInBits() == 32)
+ return false;
+
+ return true;
+}
+
+static bool isVZIPMask(const SmallVectorImpl<int> &M, EVT VT,
+ unsigned &WhichResult) {
+ unsigned NumElts = VT.getVectorNumElements();
+ WhichResult = (M[0] == 0 ? 0 : 1);
+ unsigned Idx = WhichResult * NumElts / 2;
+ for (unsigned i = 0; i != NumElts; i += 2) {
+ if ((unsigned) M[i] != Idx ||
+ (unsigned) M[i+1] != Idx + NumElts)
+ return false;
+ Idx += 1;
+ }
+
+ // VZIP.32 for 64-bit vectors is a pseudo-instruction alias for VTRN.32.
+ if (VT.is64BitVector() && VT.getVectorElementType().getSizeInBits() == 32)
+ return false;
+
+ return true;
+}
+
+static SDValue BuildSplat(SDValue Val, EVT VT, SelectionDAG &DAG, DebugLoc dl) {
// Canonicalize all-zeros and all-ones vectors.
- ConstantSDNode *ConstVal = dyn_cast<ConstantSDNode>(Val.getNode());
+ ConstantSDNode *ConstVal = cast<ConstantSDNode>(Val.getNode());
if (ConstVal->isNullValue())
return getZeroVector(VT, DAG, dl);
if (ConstVal->isAllOnesValue())
return getOnesVector(VT, DAG, dl);
- MVT CanonicalVT;
+ EVT CanonicalVT;
if (VT.is64BitVector()) {
switch (Val.getValueType().getSizeInBits()) {
case 8: CanonicalVT = MVT::v8i8; break;
// If this is a case we can't handle, return null and let the default
// expansion code take care of it.
static SDValue LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) {
- BuildVectorSDNode *BVN = dyn_cast<BuildVectorSDNode>(Op.getNode());
- assert(BVN != 0 && "Expected a BuildVectorSDNode in LowerBUILD_VECTOR");
+ BuildVectorSDNode *BVN = cast<BuildVectorSDNode>(Op.getNode());
DebugLoc dl = Op.getDebugLoc();
- MVT VT = Op.getValueType();
+ EVT VT = Op.getValueType();
APInt SplatBits, SplatUndef;
unsigned SplatBitSize;
bool HasAnyUndefs;
if (BVN->isConstantSplat(SplatBits, SplatUndef, SplatBitSize, HasAnyUndefs)) {
- SDValue Val = isVMOVSplat(SplatBits.getZExtValue(),
- SplatUndef.getZExtValue(), SplatBitSize, DAG);
- if (Val.getNode())
- return BuildSplat(Val, VT, DAG, dl);
+ if (SplatBitSize <= 64) {
+ SDValue Val = isVMOVSplat(SplatBits.getZExtValue(),
+ SplatUndef.getZExtValue(), SplatBitSize, DAG);
+ if (Val.getNode())
+ return BuildSplat(Val, VT, DAG, dl);
+ }
}
// If there are only 2 elements in a 128-bit vector, insert them into an
return SDValue();
}
-static SDValue LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) {
- return Op;
+/// isShuffleMaskLegal - Targets can use this to indicate that they only
+/// support *some* VECTOR_SHUFFLE operations, those with specific masks.
+/// By default, if a target supports the VECTOR_SHUFFLE node, all mask values
+/// are assumed to be legal.
+bool
+ARMTargetLowering::isShuffleMaskLegal(const SmallVectorImpl<int> &M,
+ EVT VT) const {
+ if (VT.getVectorNumElements() == 4 &&
+ (VT.is128BitVector() || VT.is64BitVector())) {
+ unsigned PFIndexes[4];
+ for (unsigned i = 0; i != 4; ++i) {
+ if (M[i] < 0)
+ PFIndexes[i] = 8;
+ else
+ PFIndexes[i] = M[i];
+ }
+
+ // Compute the index in the perfect shuffle table.
+ unsigned PFTableIndex =
+ PFIndexes[0]*9*9*9+PFIndexes[1]*9*9+PFIndexes[2]*9+PFIndexes[3];
+ unsigned PFEntry = PerfectShuffleTable[PFTableIndex];
+ unsigned Cost = (PFEntry >> 30);
+
+ if (Cost <= 4)
+ return true;
+ }
+
+ bool ReverseVEXT;
+ unsigned Imm, WhichResult;
+
+ return (ShuffleVectorSDNode::isSplatMask(&M[0], VT) ||
+ isVREVMask(M, VT, 64) ||
+ isVREVMask(M, VT, 32) ||
+ isVREVMask(M, VT, 16) ||
+ isVEXTMask(M, VT, ReverseVEXT, Imm) ||
+ isVTRNMask(M, VT, WhichResult) ||
+ isVUZPMask(M, VT, WhichResult) ||
+ isVZIPMask(M, VT, WhichResult));
+}
+
+/// GeneratePerfectShuffle - Given an entry in the perfect-shuffle table, emit
+/// the specified operations to build the shuffle.
+static SDValue GeneratePerfectShuffle(unsigned PFEntry, SDValue LHS,
+ SDValue RHS, SelectionDAG &DAG,
+ DebugLoc dl) {
+ unsigned OpNum = (PFEntry >> 26) & 0x0F;
+ unsigned LHSID = (PFEntry >> 13) & ((1 << 13)-1);
+ unsigned RHSID = (PFEntry >> 0) & ((1 << 13)-1);
+
+ enum {
+ OP_COPY = 0, // Copy, used for things like <u,u,u,3> to say it is <0,1,2,3>
+ OP_VREV,
+ OP_VDUP0,
+ OP_VDUP1,
+ OP_VDUP2,
+ OP_VDUP3,
+ OP_VEXT1,
+ OP_VEXT2,
+ OP_VEXT3,
+ OP_VUZPL, // VUZP, left result
+ OP_VUZPR, // VUZP, right result
+ OP_VZIPL, // VZIP, left result
+ OP_VZIPR, // VZIP, right result
+ OP_VTRNL, // VTRN, left result
+ OP_VTRNR // VTRN, right result
+ };
+
+ if (OpNum == OP_COPY) {
+ if (LHSID == (1*9+2)*9+3) return LHS;
+ assert(LHSID == ((4*9+5)*9+6)*9+7 && "Illegal OP_COPY!");
+ return RHS;
+ }
+
+ SDValue OpLHS, OpRHS;
+ OpLHS = GeneratePerfectShuffle(PerfectShuffleTable[LHSID], LHS, RHS, DAG, dl);
+ OpRHS = GeneratePerfectShuffle(PerfectShuffleTable[RHSID], LHS, RHS, DAG, dl);
+ EVT VT = OpLHS.getValueType();
+
+ switch (OpNum) {
+ default: llvm_unreachable("Unknown shuffle opcode!");
+ case OP_VREV:
+ return DAG.getNode(ARMISD::VREV64, dl, VT, OpLHS);
+ case OP_VDUP0:
+ case OP_VDUP1:
+ case OP_VDUP2:
+ case OP_VDUP3:
+ return DAG.getNode(ARMISD::VDUPLANE, dl, VT,
+ OpLHS, DAG.getConstant(OpNum-OP_VDUP0, MVT::i32));
+ case OP_VEXT1:
+ case OP_VEXT2:
+ case OP_VEXT3:
+ return DAG.getNode(ARMISD::VEXT, dl, VT,
+ OpLHS, OpRHS,
+ DAG.getConstant(OpNum-OP_VEXT1+1, MVT::i32));
+ case OP_VUZPL:
+ case OP_VUZPR:
+ return DAG.getNode(ARMISD::VUZP, dl, DAG.getVTList(VT, VT),
+ OpLHS, OpRHS).getValue(OpNum-OP_VUZPL);
+ case OP_VZIPL:
+ case OP_VZIPR:
+ return DAG.getNode(ARMISD::VZIP, dl, DAG.getVTList(VT, VT),
+ OpLHS, OpRHS).getValue(OpNum-OP_VZIPL);
+ case OP_VTRNL:
+ case OP_VTRNR:
+ return DAG.getNode(ARMISD::VTRN, dl, DAG.getVTList(VT, VT),
+ OpLHS, OpRHS).getValue(OpNum-OP_VTRNL);
+ }
}
-static SDValue LowerSCALAR_TO_VECTOR(SDValue Op, SelectionDAG &DAG) {
- return Op;
+static SDValue LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) {
+ SDValue V1 = Op.getOperand(0);
+ SDValue V2 = Op.getOperand(1);
+ DebugLoc dl = Op.getDebugLoc();
+ EVT VT = Op.getValueType();
+ ShuffleVectorSDNode *SVN = cast<ShuffleVectorSDNode>(Op.getNode());
+ SmallVector<int, 8> ShuffleMask;
+
+ // Convert shuffles that are directly supported on NEON to target-specific
+ // DAG nodes, instead of keeping them as shuffles and matching them again
+ // during code selection. This is more efficient and avoids the possibility
+ // of inconsistencies between legalization and selection.
+ // FIXME: floating-point vectors should be canonicalized to integer vectors
+ // of the same time so that they get CSEd properly.
+ SVN->getMask(ShuffleMask);
+
+ if (ShuffleVectorSDNode::isSplatMask(&ShuffleMask[0], VT)) {
+ int Lane = SVN->getSplatIndex();
+ if (Lane == 0 && V1.getOpcode() == ISD::SCALAR_TO_VECTOR) {
+ return DAG.getNode(ARMISD::VDUP, dl, VT, V1.getOperand(0));
+ }
+ return DAG.getNode(ARMISD::VDUPLANE, dl, VT, V1,
+ DAG.getConstant(Lane, MVT::i32));
+ }
+
+ bool ReverseVEXT;
+ unsigned Imm;
+ if (isVEXTMask(ShuffleMask, VT, ReverseVEXT, Imm)) {
+ if (ReverseVEXT)
+ std::swap(V1, V2);
+ return DAG.getNode(ARMISD::VEXT, dl, VT, V1, V2,
+ DAG.getConstant(Imm, MVT::i32));
+ }
+
+ if (isVREVMask(ShuffleMask, VT, 64))
+ return DAG.getNode(ARMISD::VREV64, dl, VT, V1);
+ if (isVREVMask(ShuffleMask, VT, 32))
+ return DAG.getNode(ARMISD::VREV32, dl, VT, V1);
+ if (isVREVMask(ShuffleMask, VT, 16))
+ return DAG.getNode(ARMISD::VREV16, dl, VT, V1);
+
+ // Check for Neon shuffles that modify both input vectors in place.
+ // If both results are used, i.e., if there are two shuffles with the same
+ // source operands and with masks corresponding to both results of one of
+ // these operations, DAG memoization will ensure that a single node is
+ // used for both shuffles.
+ unsigned WhichResult;
+ if (isVTRNMask(ShuffleMask, VT, WhichResult))
+ return DAG.getNode(ARMISD::VTRN, dl, DAG.getVTList(VT, VT),
+ V1, V2).getValue(WhichResult);
+ if (isVUZPMask(ShuffleMask, VT, WhichResult))
+ return DAG.getNode(ARMISD::VUZP, dl, DAG.getVTList(VT, VT),
+ V1, V2).getValue(WhichResult);
+ if (isVZIPMask(ShuffleMask, VT, WhichResult))
+ return DAG.getNode(ARMISD::VZIP, dl, DAG.getVTList(VT, VT),
+ V1, V2).getValue(WhichResult);
+
+ // If the shuffle is not directly supported and it has 4 elements, use
+ // the PerfectShuffle-generated table to synthesize it from other shuffles.
+ if (VT.getVectorNumElements() == 4 &&
+ (VT.is128BitVector() || VT.is64BitVector())) {
+ unsigned PFIndexes[4];
+ for (unsigned i = 0; i != 4; ++i) {
+ if (ShuffleMask[i] < 0)
+ PFIndexes[i] = 8;
+ else
+ PFIndexes[i] = ShuffleMask[i];
+ }
+
+ // Compute the index in the perfect shuffle table.
+ unsigned PFTableIndex =
+ PFIndexes[0]*9*9*9+PFIndexes[1]*9*9+PFIndexes[2]*9+PFIndexes[3];
+
+ unsigned PFEntry = PerfectShuffleTable[PFTableIndex];
+ unsigned Cost = (PFEntry >> 30);
+
+ if (Cost <= 4)
+ return GeneratePerfectShuffle(PFEntry, V1, V2, DAG, dl);
+ }
+
+ return SDValue();
}
static SDValue LowerEXTRACT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) {
- MVT VT = Op.getValueType();
+ EVT VT = Op.getValueType();
DebugLoc dl = Op.getDebugLoc();
- assert((VT == MVT::i8 || VT == MVT::i16) &&
- "unexpected type for custom-lowering vector extract");
SDValue Vec = Op.getOperand(0);
SDValue Lane = Op.getOperand(1);
- Op = DAG.getNode(ARMISD::VGETLANEu, dl, MVT::i32, Vec, Lane);
- Op = DAG.getNode(ISD::AssertZext, dl, MVT::i32, Op, DAG.getValueType(VT));
- return DAG.getNode(ISD::TRUNCATE, dl, VT, Op);
+ assert(VT == MVT::i32 &&
+ Vec.getValueType().getVectorElementType().getSizeInBits() < 32 &&
+ "unexpected type for custom-lowering vector extract");
+ return DAG.getNode(ARMISD::VGETLANEu, dl, MVT::i32, Vec, Lane);
}
static SDValue LowerCONCAT_VECTORS(SDValue Op, SelectionDAG &DAG) {
case ISD::RETURNADDR: break;
case ISD::FRAMEADDR: return LowerFRAMEADDR(Op, DAG);
case ISD::GLOBAL_OFFSET_TABLE: return LowerGLOBAL_OFFSET_TABLE(Op, DAG);
- case ISD::INTRINSIC_VOID:
- case ISD::INTRINSIC_W_CHAIN: return LowerINTRINSIC_W_CHAIN(Op, DAG);
case ISD::INTRINSIC_WO_CHAIN: return LowerINTRINSIC_WO_CHAIN(Op, DAG);
case ISD::BIT_CONVERT: return ExpandBIT_CONVERT(Op.getNode(), DAG);
case ISD::SHL:
case ISD::VSETCC: return LowerVSETCC(Op, DAG);
case ISD::BUILD_VECTOR: return LowerBUILD_VECTOR(Op, DAG);
case ISD::VECTOR_SHUFFLE: return LowerVECTOR_SHUFFLE(Op, DAG);
- case ISD::SCALAR_TO_VECTOR: return LowerSCALAR_TO_VECTOR(Op, DAG);
case ISD::EXTRACT_VECTOR_ELT: return LowerEXTRACT_VECTOR_ELT(Op, DAG);
case ISD::CONCAT_VECTORS: return LowerCONCAT_VECTORS(Op, DAG);
}
MachineBasicBlock *
ARMTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
- MachineBasicBlock *BB) const {
+ MachineBasicBlock *BB,
+ DenseMap<MachineBasicBlock*, MachineBasicBlock*> *EM) const {
const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
DebugLoc dl = MI->getDebugLoc();
switch (MI->getOpcode()) {
default:
llvm_unreachable("Unexpected instr type to insert");
- case ARM::tMOVCCr: {
+ case ARM::tMOVCCr_pseudo: {
// To "insert" a SELECT_CC instruction, we actually have to insert the
// diamond control-flow pattern. The incoming instruction knows the
// destination vreg to set, the condition code register to branch on, the
F->insert(It, sinkMBB);
// Update machine-CFG edges by first adding all successors of the current
// block to the new block which will contain the Phi node for the select.
- for(MachineBasicBlock::succ_iterator i = BB->succ_begin(),
- e = BB->succ_end(); i != e; ++i)
- sinkMBB->addSuccessor(*i);
+ // Also inform sdisel of the edge changes.
+ for (MachineBasicBlock::succ_iterator I = BB->succ_begin(),
+ E = BB->succ_end(); I != E; ++I) {
+ EM->insert(std::make_pair(*I, sinkMBB));
+ sinkMBB->addSuccessor(*I);
+ }
// Next, remove all successors of the current block, and add the true
// and fallthrough blocks as its successors.
- while(!BB->succ_empty())
+ while (!BB->succ_empty())
BB->removeSuccessor(BB->succ_begin());
BB->addSuccessor(copy0MBB);
BB->addSuccessor(sinkMBB);
TargetLowering::DAGCombinerInfo &DCI) {
SelectionDAG &DAG = DCI.DAG;
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
- MVT VT = N->getValueType(0);
+ EVT VT = N->getValueType(0);
unsigned Opc = N->getOpcode();
bool isSlctCC = Slct.getOpcode() == ISD::SELECT_CC;
SDValue LHS = isSlctCC ? Slct.getOperand(2) : Slct.getOperand(1);
cast<ConstantSDNode>(RHS)->isNullValue()) {
std::swap(LHS, RHS);
SDValue Op0 = Slct.getOperand(0);
- MVT OpVT = isSlctCC ? Op0.getValueType() :
+ EVT OpVT = isSlctCC ? Op0.getValueType() :
Op0.getOperand(0).getValueType();
bool isInt = OpVT.isInteger();
CC = ISD::getSetCCInverse(CC, isInt);
/// operand of a vector shift left operation. That value must be in the range:
/// 0 <= Value < ElementBits for a left shift; or
/// 0 <= Value <= ElementBits for a long left shift.
-static bool isVShiftLImm(SDValue Op, MVT VT, bool isLong, int64_t &Cnt) {
+static bool isVShiftLImm(SDValue Op, EVT VT, bool isLong, int64_t &Cnt) {
assert(VT.isVector() && "vector shift count is not a vector type");
unsigned ElementBits = VT.getVectorElementType().getSizeInBits();
if (! getVShiftImm(Op, ElementBits, Cnt))
/// absolute value must be in the range:
/// 1 <= |Value| <= ElementBits for a right shift; or
/// 1 <= |Value| <= ElementBits/2 for a narrow right shift.
-static bool isVShiftRImm(SDValue Op, MVT VT, bool isNarrow, bool isIntrinsic,
+static bool isVShiftRImm(SDValue Op, EVT VT, bool isNarrow, bool isIntrinsic,
int64_t &Cnt) {
assert(VT.isVector() && "vector shift count is not a vector type");
unsigned ElementBits = VT.getVectorElementType().getSizeInBits();
case Intrinsic::arm_neon_vqrshiftns:
case Intrinsic::arm_neon_vqrshiftnu:
case Intrinsic::arm_neon_vqrshiftnsu: {
- MVT VT = N->getOperand(1).getValueType();
+ EVT VT = N->getOperand(1).getValueType();
int64_t Cnt;
unsigned VShiftOpc = 0;
}
case Intrinsic::arm_neon_vshiftins: {
- MVT VT = N->getOperand(1).getValueType();
+ EVT VT = N->getOperand(1).getValueType();
int64_t Cnt;
unsigned VShiftOpc = 0;
/// their values after they get legalized to loads from a constant pool.
static SDValue PerformShiftCombine(SDNode *N, SelectionDAG &DAG,
const ARMSubtarget *ST) {
- MVT VT = N->getValueType(0);
+ EVT VT = N->getValueType(0);
// Nothing to be done for scalar shifts.
if (! VT.isVector())
if (ST->hasNEON() && N0.getOpcode() == ISD::EXTRACT_VECTOR_ELT) {
SDValue Vec = N0.getOperand(0);
SDValue Lane = N0.getOperand(1);
- MVT VT = N->getValueType(0);
- MVT EltVT = N0.getValueType();
+ EVT VT = N->getValueType(0);
+ EVT EltVT = N0.getValueType();
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
if (VT == MVT::i32 &&
return SDValue();
}
+bool ARMTargetLowering::allowsUnalignedMemoryAccesses(EVT VT) const {
+ if (!Subtarget->hasV6Ops())
+ // Pre-v6 does not support unaligned mem access.
+ return false;
+ else if (!Subtarget->hasV6Ops()) {
+ // v6 may or may not support unaligned mem access.
+ if (!Subtarget->isTargetDarwin())
+ return false;
+ }
+
+ switch (VT.getSimpleVT().SimpleTy) {
+ default:
+ return false;
+ case MVT::i8:
+ case MVT::i16:
+ case MVT::i32:
+ return true;
+ // FIXME: VLD1 etc with standard alignment is legal.
+ }
+}
+
+static bool isLegalT1AddressImmediate(int64_t V, EVT VT) {
+ if (V < 0)
+ return false;
+
+ unsigned Scale = 1;
+ switch (VT.getSimpleVT().SimpleTy) {
+ default: return false;
+ case MVT::i1:
+ case MVT::i8:
+ // Scale == 1;
+ break;
+ case MVT::i16:
+ // Scale == 2;
+ Scale = 2;
+ break;
+ case MVT::i32:
+ // Scale == 4;
+ Scale = 4;
+ break;
+ }
+
+ if ((V & (Scale - 1)) != 0)
+ return false;
+ V /= Scale;
+ return V == (V & ((1LL << 5) - 1));
+}
+
+static bool isLegalT2AddressImmediate(int64_t V, EVT VT,
+ const ARMSubtarget *Subtarget) {
+ bool isNeg = false;
+ if (V < 0) {
+ isNeg = true;
+ V = - V;
+ }
+
+ switch (VT.getSimpleVT().SimpleTy) {
+ default: return false;
+ case MVT::i1:
+ case MVT::i8:
+ case MVT::i16:
+ case MVT::i32:
+ // + imm12 or - imm8
+ if (isNeg)
+ return V == (V & ((1LL << 8) - 1));
+ return V == (V & ((1LL << 12) - 1));
+ case MVT::f32:
+ case MVT::f64:
+ // Same as ARM mode. FIXME: NEON?
+ if (!Subtarget->hasVFP2())
+ return false;
+ if ((V & 3) != 0)
+ return false;
+ V >>= 2;
+ return V == (V & ((1LL << 8) - 1));
+ }
+}
+
/// isLegalAddressImmediate - Return true if the integer value can be used
/// as the offset of the target addressing mode for load / store of the
/// given type.
-static bool isLegalAddressImmediate(int64_t V, MVT VT,
+static bool isLegalAddressImmediate(int64_t V, EVT VT,
const ARMSubtarget *Subtarget) {
if (V == 0)
return true;
if (!VT.isSimple())
return false;
- if (Subtarget->isThumb()) { // FIXME for thumb2
- if (V < 0)
- return false;
-
- unsigned Scale = 1;
- switch (VT.getSimpleVT()) {
- default: return false;
- case MVT::i1:
- case MVT::i8:
- // Scale == 1;
- break;
- case MVT::i16:
- // Scale == 2;
- Scale = 2;
- break;
- case MVT::i32:
- // Scale == 4;
- Scale = 4;
- break;
- }
-
- if ((V & (Scale - 1)) != 0)
- return false;
- V /= Scale;
- return V == (V & ((1LL << 5) - 1));
- }
+ if (Subtarget->isThumb1Only())
+ return isLegalT1AddressImmediate(V, VT);
+ else if (Subtarget->isThumb2())
+ return isLegalT2AddressImmediate(V, VT, Subtarget);
+ // ARM mode.
if (V < 0)
V = - V;
- switch (VT.getSimpleVT()) {
+ switch (VT.getSimpleVT().SimpleTy) {
default: return false;
case MVT::i1:
case MVT::i8:
return V == (V & ((1LL << 8) - 1));
case MVT::f32:
case MVT::f64:
- if (!Subtarget->hasVFP2())
+ if (!Subtarget->hasVFP2()) // FIXME: NEON?
return false;
if ((V & 3) != 0)
return false;
}
}
+bool ARMTargetLowering::isLegalT2ScaledAddressingMode(const AddrMode &AM,
+ EVT VT) const {
+ int Scale = AM.Scale;
+ if (Scale < 0)
+ return false;
+
+ switch (VT.getSimpleVT().SimpleTy) {
+ default: return false;
+ case MVT::i1:
+ case MVT::i8:
+ case MVT::i16:
+ case MVT::i32:
+ if (Scale == 1)
+ return true;
+ // r + r << imm
+ Scale = Scale & ~1;
+ return Scale == 2 || Scale == 4 || Scale == 8;
+ case MVT::i64:
+ // r + r
+ if (((unsigned)AM.HasBaseReg + Scale) <= 2)
+ return true;
+ return false;
+ case MVT::isVoid:
+ // Note, we allow "void" uses (basically, uses that aren't loads or
+ // stores), because arm allows folding a scale into many arithmetic
+ // operations. This should be made more precise and revisited later.
+
+ // Allow r << imm, but the imm has to be a multiple of two.
+ if (Scale & 1) return false;
+ return isPowerOf2_32(Scale);
+ }
+}
+
/// isLegalAddressingMode - Return true if the addressing mode represented
/// by AM is legal for this target, for a load/store of the specified type.
bool ARMTargetLowering::isLegalAddressingMode(const AddrMode &AM,
const Type *Ty) const {
- MVT VT = getValueType(Ty, true);
+ EVT VT = getValueType(Ty, true);
if (!isLegalAddressImmediate(AM.BaseOffs, VT, Subtarget))
return false;
case 0: // no scale reg, must be "r+i" or "r", or "i".
break;
case 1:
- if (Subtarget->isThumb()) // FIXME for thumb2
+ if (Subtarget->isThumb1Only())
return false;
// FALL THROUGH.
default:
if (!VT.isSimple())
return false;
+ if (Subtarget->isThumb2())
+ return isLegalT2ScaledAddressingMode(AM, VT);
+
int Scale = AM.Scale;
- switch (VT.getSimpleVT()) {
+ switch (VT.getSimpleVT().SimpleTy) {
default: return false;
case MVT::i1:
case MVT::i8:
case MVT::i32:
- case MVT::i64:
- // This assumes i64 is legalized to a pair of i32. If not (i.e.
- // ldrd / strd are used, then its address mode is same as i16.
- // r + r
if (Scale < 0) Scale = -Scale;
if (Scale == 1)
return true;
// r + r << imm
return isPowerOf2_32(Scale & ~1);
case MVT::i16:
+ case MVT::i64:
// r + r
if (((unsigned)AM.HasBaseReg + Scale) <= 2)
return true;
// operations. This should be made more precise and revisited later.
// Allow r << imm, but the imm has to be a multiple of two.
- if (AM.Scale & 1) return false;
- return isPowerOf2_32(AM.Scale);
+ if (Scale & 1) return false;
+ return isPowerOf2_32(Scale);
}
break;
}
return true;
}
-static bool getARMIndexedAddressParts(SDNode *Ptr, MVT VT,
+static bool getARMIndexedAddressParts(SDNode *Ptr, EVT VT,
bool isSEXTLoad, SDValue &Base,
SDValue &Offset, bool &isInc,
SelectionDAG &DAG) {
return false;
}
-static bool getT2IndexedAddressParts(SDNode *Ptr, MVT VT,
+static bool getT2IndexedAddressParts(SDNode *Ptr, EVT VT,
bool isSEXTLoad, SDValue &Base,
SDValue &Offset, bool &isInc,
SelectionDAG &DAG) {
if (Subtarget->isThumb1Only())
return false;
- MVT VT;
+ EVT VT;
SDValue Ptr;
bool isSEXTLoad = false;
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
bool isInc;
bool isLegal = false;
- if (Subtarget->isThumb() && Subtarget->hasThumb2())
+ if (Subtarget->isThumb2())
isLegal = getT2IndexedAddressParts(Ptr.getNode(), VT, isSEXTLoad, Base,
Offset, isInc, DAG);
- else
+ else
isLegal = getARMIndexedAddressParts(Ptr.getNode(), VT, isSEXTLoad, Base,
Offset, isInc, DAG);
if (!isLegal)
if (Subtarget->isThumb1Only())
return false;
- MVT VT;
+ EVT VT;
SDValue Ptr;
bool isSEXTLoad = false;
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
bool isInc;
bool isLegal = false;
- if (Subtarget->isThumb() && Subtarget->hasThumb2())
+ if (Subtarget->isThumb2())
isLegal = getT2IndexedAddressParts(Op, VT, isSEXTLoad, Base, Offset,
isInc, DAG);
- else
+ else
isLegal = getARMIndexedAddressParts(Op, VT, isSEXTLoad, Base, Offset,
isInc, DAG);
if (!isLegal)
std::pair<unsigned, const TargetRegisterClass*>
ARMTargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint,
- MVT VT) const {
+ EVT VT) const {
if (Constraint.size() == 1) {
// GCC RS6000 Constraint Letters
switch (Constraint[0]) {
std::vector<unsigned> ARMTargetLowering::
getRegClassForInlineAsmConstraint(const std::string &Constraint,
- MVT VT) const {
+ EVT VT) const {
if (Constraint.size() != 1)
return std::vector<unsigned>();
return TargetLowering::LowerAsmOperandForConstraint(Op, Constraint, hasMemory,
Ops, DAG);
}
+
+bool
+ARMTargetLowering::isOffsetFoldingLegal(const GlobalAddressSDNode *GA) const {
+ // The ARM target isn't yet aware of offsets.
+ return false;
+}