#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineModuleInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/PseudoSourceValue.h"
#include "llvm/CodeGen/SelectionDAG.h"
EVT ElemTy = VT.getVectorElementType();
if (ElemTy != MVT::i64 && ElemTy != MVT::f64)
setOperationAction(ISD::SETCC, VT.getSimpleVT(), Custom);
+ setOperationAction(ISD::INSERT_VECTOR_ELT, VT.getSimpleVT(), Custom);
setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT.getSimpleVT(), Custom);
- if (ElemTy != MVT::i32) {
+ if (ElemTy == MVT::i32) {
+ setOperationAction(ISD::SINT_TO_FP, VT.getSimpleVT(), Custom);
+ setOperationAction(ISD::UINT_TO_FP, VT.getSimpleVT(), Custom);
+ setOperationAction(ISD::FP_TO_SINT, VT.getSimpleVT(), Custom);
+ setOperationAction(ISD::FP_TO_UINT, VT.getSimpleVT(), Custom);
+ } else {
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::EXTRACT_SUBVECTOR, VT.getSimpleVT(), Legal);
setOperationAction(ISD::SELECT, VT.getSimpleVT(), Expand);
setOperationAction(ISD::SELECT_CC, VT.getSimpleVT(), Expand);
+ setOperationAction(ISD::SIGN_EXTEND_INREG, VT.getSimpleVT(), Expand);
if (VT.isInteger()) {
setOperationAction(ISD::SHL, VT.getSimpleVT(), Custom);
setOperationAction(ISD::SRA, VT.getSimpleVT(), Custom);
setOperationAction(ISD::SRL, VT.getSimpleVT(), Custom);
- setLoadExtAction(ISD::SEXTLOAD, VT.getSimpleVT(), Expand);
- setLoadExtAction(ISD::ZEXTLOAD, VT.getSimpleVT(), Expand);
- for (unsigned InnerVT = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
- InnerVT <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++InnerVT)
- setTruncStoreAction(VT.getSimpleVT(),
- (MVT::SimpleValueType)InnerVT, Expand);
}
- setLoadExtAction(ISD::EXTLOAD, VT.getSimpleVT(), Expand);
// Promote all bit-wise operations.
if (VT.isInteger() && VT != PromotedBitwiseVT) {
setLibcallName(RTLIB::MEMSET, "__aeabi_memset");
}
+ // Use divmod compiler-rt calls for iOS 5.0 and later.
+ if (Subtarget->getTargetTriple().getOS() == Triple::IOS &&
+ !Subtarget->getTargetTriple().isOSVersionLT(5, 0)) {
+ setLibcallName(RTLIB::SDIVREM_I32, "__divmodsi4");
+ setLibcallName(RTLIB::UDIVREM_I32, "__udivmodsi4");
+ }
+
if (Subtarget->isThumb1Only())
addRegisterClass(MVT::i32, ARM::tGPRRegisterClass);
else
setTruncStoreAction(MVT::f64, MVT::f32, Expand);
}
+ for (unsigned VT = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
+ VT <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++VT) {
+ for (unsigned InnerVT = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
+ InnerVT <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++InnerVT)
+ setTruncStoreAction((MVT::SimpleValueType)VT,
+ (MVT::SimpleValueType)InnerVT, Expand);
+ setLoadExtAction(ISD::SEXTLOAD, (MVT::SimpleValueType)VT, Expand);
+ setLoadExtAction(ISD::ZEXTLOAD, (MVT::SimpleValueType)VT, Expand);
+ setLoadExtAction(ISD::EXTLOAD, (MVT::SimpleValueType)VT, Expand);
+ }
+
if (Subtarget->hasNEON()) {
addDRTypeForNEON(MVT::v2f32);
addDRTypeForNEON(MVT::v8i8);
setOperationAction(ISD::FNEARBYINT, MVT::v2f64, Expand);
setOperationAction(ISD::FFLOOR, MVT::v2f64, Expand);
- setTruncStoreAction(MVT::v2f64, MVT::v2f32, Expand);
-
// Neon does not support some operations on v1i64 and v2i64 types.
setOperationAction(ISD::MUL, MVT::v1i64, Expand);
// Custom handling for some quad-vector types to detect VMULL.
setTargetDAGCombine(ISD::FP_TO_SINT);
setTargetDAGCombine(ISD::FP_TO_UINT);
setTargetDAGCombine(ISD::FDIV);
+
+ setLoadExtAction(ISD::EXTLOAD, MVT::v4i8, Expand);
}
computeRegisterProperties();
//// temporary - rewrite interface to use type
maxStoresPerMemcpy = maxStoresPerMemcpyOptSize = 1;
+ maxStoresPerMemset = 16;
+ maxStoresPerMemsetOptSize = Subtarget->isTargetDarwin() ? 8 : 4;
// On ARM arguments smaller than 4 bytes are extended, so all arguments
// are at least 4 bytes aligned.
if (VT == MVT::Glue || VT == MVT::Other)
continue;
if (VT.isFloatingPoint() || VT.isVector())
- return Sched::Latency;
+ return Sched::ILP;
}
if (!N->isMachineOpcode())
return Sched::RegPressure;
if (!Itins->isEmpty() &&
Itins->getOperandCycle(MCID.getSchedClass(), 0) > 2)
- return Sched::Latency;
+ return Sched::ILP;
return Sched::RegPressure;
}
MachineFunction &MF = DAG.getMachineFunction();
bool IsStructRet = (Outs.empty()) ? false : Outs[0].Flags.isSRet();
bool IsSibCall = false;
- // Temporarily disable tail calls so things don't break.
- if (!EnableARMTailCalls)
+ // Disable tail calls if they're not supported.
+ if (!EnableARMTailCalls && !Subtarget->supportsTailCall())
isTailCall = false;
if (isTailCall) {
// Check if it's really possible to do a tail call.
SDValue AddArg = DAG.getNode(ISD::ADD, dl, PtrVT, Arg, Const);
SDValue Load = DAG.getLoad(PtrVT, dl, Chain, AddArg,
MachinePointerInfo(),
- false, false, 0);
+ false, false, false, 0);
MemOpChains.push_back(Load.getValue(1));
RegsToPass.push_back(std::make_pair(j, Load));
}
SDValue Src = DAG.getNode(ISD::ADD, dl, getPointerTy(), Arg, SrcOffset);
SDValue SizeNode = DAG.getConstant(Flags.getByValSize() - 4*offset,
MVT::i32);
- // TODO: Disable AlwaysInline when it becomes possible
- // to emit a nested call sequence.
MemOpChains.push_back(DAG.getMemcpy(Chain, dl, Dst, Src, SizeNode,
Flags.getByValAlign(),
/*isVolatile=*/false,
- /*AlwaysInline=*/true,
+ /*AlwaysInline=*/false,
MachinePointerInfo(0),
MachinePointerInfo(0)));
Callee = DAG.getLoad(getPointerTy(), dl,
DAG.getEntryNode(), CPAddr,
MachinePointerInfo::getConstantPool(),
- false, false, 0);
+ false, false, false, 0);
} else if (ExternalSymbolSDNode *S=dyn_cast<ExternalSymbolSDNode>(Callee)) {
const char *Sym = S->getSymbol();
// Create a constant pool entry for the callee address
unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
- ARMConstantPoolValue *CPV = new ARMConstantPoolValue(*DAG.getContext(),
- Sym, ARMPCLabelIndex, 0);
+ ARMConstantPoolValue *CPV =
+ ARMConstantPoolSymbol::Create(*DAG.getContext(), Sym,
+ ARMPCLabelIndex, 0);
// Get the address of the callee into a register
SDValue CPAddr = DAG.getTargetConstantPool(CPV, getPointerTy(), 4);
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
Callee = DAG.getLoad(getPointerTy(), dl,
DAG.getEntryNode(), CPAddr,
MachinePointerInfo::getConstantPool(),
- false, false, 0);
+ false, false, false, 0);
}
} else if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
const GlobalValue *GV = G->getGlobal();
Callee = DAG.getLoad(getPointerTy(), dl,
DAG.getEntryNode(), CPAddr,
MachinePointerInfo::getConstantPool(),
- false, false, 0);
+ false, false, false, 0);
SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
Callee = DAG.getNode(ARMISD::PIC_ADD, dl,
getPointerTy(), Callee, PICLabel);
const char *Sym = S->getSymbol();
if (isARMFunc && Subtarget->isThumb1Only() && !Subtarget->hasV5TOps()) {
unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
- ARMConstantPoolValue *CPV = new ARMConstantPoolValue(*DAG.getContext(),
- Sym, ARMPCLabelIndex, 4);
+ ARMConstantPoolValue *CPV =
+ ARMConstantPoolSymbol::Create(*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,
DAG.getEntryNode(), CPAddr,
MachinePointerInfo::getConstantPool(),
- false, false, 0);
+ false, false, false, 0);
SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
Callee = DAG.getNode(ARMISD::PIC_ADD, dl,
getPointerTy(), Callee, PICLabel);
CPAddr = DAG.getNode(ARMISD::Wrapper, DL, PtrVT, CPAddr);
SDValue Result = DAG.getLoad(PtrVT, DL, DAG.getEntryNode(), CPAddr,
MachinePointerInfo::getConstantPool(),
- false, false, 0);
+ false, false, false, 0);
if (RelocM == Reloc::Static)
return Result;
SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
Argument = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, Argument);
Argument = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), Argument,
MachinePointerInfo::getConstantPool(),
- false, false, 0);
+ false, false, false, 0);
SDValue Chain = Argument.getValue(1);
SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
Offset = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, Offset);
Offset = DAG.getLoad(PtrVT, dl, Chain, Offset,
MachinePointerInfo::getConstantPool(),
- false, false, 0);
+ false, false, false, 0);
Chain = Offset.getValue(1);
SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
Offset = DAG.getLoad(PtrVT, dl, Chain, Offset,
MachinePointerInfo::getConstantPool(),
- false, false, 0);
+ false, false, false, 0);
} else {
// local exec model
ARMConstantPoolValue *CPV =
Offset = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, Offset);
Offset = DAG.getLoad(PtrVT, dl, Chain, Offset,
MachinePointerInfo::getConstantPool(),
- false, false, 0);
+ false, false, false, 0);
}
// The address of the thread local variable is the add of the thread
SDValue Result = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(),
CPAddr,
MachinePointerInfo::getConstantPool(),
- false, false, 0);
+ false, false, false, 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,
- MachinePointerInfo::getGOT(), false, false, 0);
+ MachinePointerInfo::getGOT(),
+ false, false, false, 0);
return Result;
}
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
return DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), CPAddr,
MachinePointerInfo::getConstantPool(),
- false, false, 0);
+ false, false, false, 0);
}
}
DAG.getTargetGlobalAddress(GV, dl, PtrVT));
if (Subtarget->GVIsIndirectSymbol(GV, RelocM))
Result = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), Result,
- MachinePointerInfo::getGOT(), false, false, 0);
+ MachinePointerInfo::getGOT(),
+ false, false, false, 0);
return Result;
}
SDValue Result = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), CPAddr,
MachinePointerInfo::getConstantPool(),
- false, false, 0);
+ false, false, false, 0);
SDValue Chain = Result.getValue(1);
if (RelocM == Reloc::PIC_) {
if (Subtarget->GVIsIndirectSymbol(GV, RelocM))
Result = DAG.getLoad(PtrVT, dl, Chain, Result, MachinePointerInfo::getGOT(),
- false, false, 0);
+ false, false, false, 0);
return Result;
}
EVT PtrVT = getPointerTy();
DebugLoc dl = Op.getDebugLoc();
unsigned PCAdj = Subtarget->isThumb() ? 4 : 8;
- ARMConstantPoolValue *CPV = new ARMConstantPoolValue(*DAG.getContext(),
- "_GLOBAL_OFFSET_TABLE_",
- ARMPCLabelIndex, PCAdj);
+ ARMConstantPoolValue *CPV =
+ ARMConstantPoolSymbol::Create(*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,
MachinePointerInfo::getConstantPool(),
- false, false, 0);
+ false, false, false, 0);
SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
return DAG.getNode(ARMISD::PIC_ADD, dl, PtrVT, Result, PICLabel);
}
ARMTargetLowering::LowerEH_SJLJ_SETJMP(SDValue Op, SelectionDAG &DAG) const {
DebugLoc dl = Op.getDebugLoc();
SDValue Val = DAG.getConstant(0, MVT::i32);
- return DAG.getNode(ARMISD::EH_SJLJ_SETJMP, dl, MVT::i32, Op.getOperand(0),
+ return DAG.getNode(ARMISD::EH_SJLJ_SETJMP, dl,
+ DAG.getVTList(MVT::i32, MVT::Other), Op.getOperand(0),
Op.getOperand(1), Val);
}
SDValue Result =
DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), CPAddr,
MachinePointerInfo::getConstantPool(),
- false, false, 0);
+ false, false, false, 0);
if (RelocM == Reloc::PIC_) {
SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
SDValue FIN = DAG.getFrameIndex(FI, getPointerTy());
ArgValue2 = DAG.getLoad(MVT::i32, dl, Root, FIN,
MachinePointerInfo::getFixedStack(FI),
- false, false, 0);
+ false, false, false, 0);
} else {
Reg = MF.addLiveIn(NextVA.getLocReg(), RC);
ArgValue2 = DAG.getCopyFromReg(Root, dl, Reg, MVT::i32);
SDValue FIN = DAG.getFrameIndex(FI, getPointerTy());
ArgValue2 = DAG.getLoad(MVT::f64, dl, Chain, FIN,
MachinePointerInfo::getFixedStack(FI),
- false, false, 0);
+ false, false, false, 0);
} else {
ArgValue2 = GetF64FormalArgument(VA, ArgLocs[++i],
Chain, DAG, dl);
SDValue FIN = DAG.getFrameIndex(FI, getPointerTy());
InVals.push_back(DAG.getLoad(VA.getValVT(), dl, Chain, FIN,
MachinePointerInfo::getFixedStack(FI),
- false, false, 0));
+ false, false, false, 0));
}
lastInsIndex = index;
}
return DAG.getLoad(MVT::i32, Op.getDebugLoc(),
Ld->getChain(), Ld->getBasePtr(), Ld->getPointerInfo(),
Ld->isVolatile(), Ld->isNonTemporal(),
- Ld->getAlignment());
+ Ld->isInvariant(), Ld->getAlignment());
llvm_unreachable("Unknown VFP cmp argument!");
}
Ld->getChain(), Ptr,
Ld->getPointerInfo(),
Ld->isVolatile(), Ld->isNonTemporal(),
- Ld->getAlignment());
+ Ld->isInvariant(), Ld->getAlignment());
EVT PtrType = Ptr.getValueType();
unsigned NewAlign = MinAlign(Ld->getAlignment(), 4);
Ld->getChain(), NewPtr,
Ld->getPointerInfo().getWithOffset(4),
Ld->isVolatile(), Ld->isNonTemporal(),
- NewAlign);
+ Ld->isInvariant(), NewAlign);
return;
}
if (getTargetMachine().getRelocationModel() == Reloc::PIC_) {
Addr = DAG.getLoad((EVT)MVT::i32, dl, Chain, Addr,
MachinePointerInfo::getJumpTable(),
- false, false, 0);
+ false, false, false, 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);
} else {
Addr = DAG.getLoad(PTy, dl, Chain, Addr,
- MachinePointerInfo::getJumpTable(), false, false, 0);
+ MachinePointerInfo::getJumpTable(),
+ false, false, false, 0);
Chain = Addr.getValue(1);
return DAG.getNode(ARMISD::BR_JT, dl, MVT::Other, Chain, Addr, JTI, UId);
}
}
+static SDValue LowerVectorFP_TO_INT(SDValue Op, SelectionDAG &DAG) {
+ EVT VT = Op.getValueType();
+ assert(VT.getVectorElementType() == MVT::i32 && "Unexpected custom lowering");
+
+ if (Op.getOperand(0).getValueType().getVectorElementType() == MVT::f32)
+ return Op;
+ return DAG.UnrollVectorOp(Op.getNode());
+}
+
static SDValue LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG) {
+ EVT VT = Op.getValueType();
+ if (VT.isVector())
+ return LowerVectorFP_TO_INT(Op, DAG);
+
DebugLoc dl = Op.getDebugLoc();
unsigned Opc;
EVT VT = Op.getValueType();
DebugLoc dl = Op.getDebugLoc();
+ if (Op.getOperand(0).getValueType().getVectorElementType() == MVT::i32) {
+ if (VT.getVectorElementType() == MVT::f32)
+ return Op;
+ return DAG.UnrollVectorOp(Op.getNode());
+ }
+
assert(Op.getOperand(0).getValueType() == MVT::v4i16 &&
"Invalid type for custom lowering!");
if (VT != MVT::v4f32)
SDValue Offset = DAG.getConstant(4, MVT::i32);
return DAG.getLoad(VT, dl, DAG.getEntryNode(),
DAG.getNode(ISD::ADD, dl, VT, FrameAddr, Offset),
- MachinePointerInfo(), false, false, 0);
+ MachinePointerInfo(), false, false, false, 0);
}
// Return LR, which contains the return address. Mark it an implicit live-in.
while (Depth--)
FrameAddr = DAG.getLoad(VT, dl, DAG.getEntryNode(), FrameAddr,
MachinePointerInfo(),
- false, false, 0);
+ false, false, false, 0);
return FrameAddr;
}
}
// Try an immediate VMVN.
- uint64_t NegatedImm = (SplatBits.getZExtValue() ^
- ((1LL << SplatBitSize) - 1));
+ uint64_t NegatedImm = (~SplatBits).getZExtValue();
Val = isNEONModifiedImm(NegatedImm,
SplatUndef.getZExtValue(), SplatBitSize,
DAG, VmovVT, VT.is128BitVector(),
// A shuffle can only come from building a vector from various
// elements of other vectors.
return SDValue();
+ } else if (V.getOperand(0).getValueType().getVectorElementType() !=
+ VT.getVectorElementType()) {
+ // This code doesn't know how to handle shuffles where the vector
+ // element types do not match (this happens because type legalization
+ // promotes the return type of EXTRACT_VECTOR_ELT).
+ // FIXME: It might be appropriate to extend this code to handle
+ // mismatched types.
+ return SDValue();
}
// Record this extraction against the appropriate vector if possible...
// If this is undef splat, generate it via "just" vdup, if possible.
if (Lane == -1) Lane = 0;
+ // Test if V1 is a SCALAR_TO_VECTOR.
if (Lane == 0 && V1.getOpcode() == ISD::SCALAR_TO_VECTOR) {
return DAG.getNode(ARMISD::VDUP, dl, VT, V1.getOperand(0));
}
+ // Test if V1 is a BUILD_VECTOR which is equivalent to a SCALAR_TO_VECTOR
+ // (and probably will turn into a SCALAR_TO_VECTOR once legalization
+ // reaches it).
+ if (Lane == 0 && V1.getOpcode() == ISD::BUILD_VECTOR &&
+ !isa<ConstantSDNode>(V1.getOperand(0))) {
+ bool IsScalarToVector = true;
+ for (unsigned i = 1, e = V1.getNumOperands(); i != e; ++i)
+ if (V1.getOperand(i).getOpcode() != ISD::UNDEF) {
+ IsScalarToVector = false;
+ break;
+ }
+ if (IsScalarToVector)
+ return DAG.getNode(ARMISD::VDUP, dl, VT, V1.getOperand(0));
+ }
return DAG.getNode(ARMISD::VDUPLANE, dl, VT, V1,
DAG.getConstant(Lane, MVT::i32));
}
return SDValue();
}
+static SDValue LowerINSERT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) {
+ // INSERT_VECTOR_ELT is legal only for immediate indexes.
+ SDValue Lane = Op.getOperand(2);
+ if (!isa<ConstantSDNode>(Lane))
+ return SDValue();
+
+ return Op;
+}
+
static SDValue LowerEXTRACT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) {
// EXTRACT_VECTOR_ELT is legal only for immediate indexes.
SDValue Lane = Op.getOperand(1);
unsigned EltSize = VT.getVectorElementType().getSizeInBits();
unsigned HalfSize = EltSize / 2;
if (isSigned) {
- int64_t SExtVal = C->getSExtValue();
- if ((SExtVal >> HalfSize) != (SExtVal >> EltSize))
+ if (!isIntN(HalfSize, C->getSExtValue()))
return false;
} else {
- if ((C->getZExtValue() >> HalfSize) != 0)
+ if (!isUIntN(HalfSize, C->getZExtValue()))
return false;
}
continue;
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N))
return DAG.getLoad(LD->getMemoryVT(), N->getDebugLoc(), LD->getChain(),
LD->getBasePtr(), LD->getPointerInfo(), LD->isVolatile(),
- LD->isNonTemporal(), LD->getAlignment());
+ LD->isNonTemporal(), LD->isInvariant(),
+ LD->getAlignment());
// Otherwise, the value must be a BUILD_VECTOR. For v2i64, it will
// have been legalized as a BITCAST from v4i32.
if (N->getOpcode() == ISD::BITCAST) {
static void
ReplaceATOMIC_OP_64(SDNode *Node, SmallVectorImpl<SDValue>& Results,
SelectionDAG &DAG, unsigned NewOp) {
- EVT T = Node->getValueType(0);
DebugLoc dl = Node->getDebugLoc();
- assert (T == MVT::i64 && "Only know how to expand i64 atomics");
+ assert (Node->getValueType(0) == MVT::i64 &&
+ "Only know how to expand i64 atomics");
SmallVector<SDValue, 6> Ops;
Ops.push_back(Node->getOperand(0)); // Chain
case ISD::SETCC: return LowerVSETCC(Op, DAG);
case ISD::BUILD_VECTOR: return LowerBUILD_VECTOR(Op, DAG, Subtarget);
case ISD::VECTOR_SHUFFLE: return LowerVECTOR_SHUFFLE(Op, DAG);
+ case ISD::INSERT_VECTOR_ELT: return LowerINSERT_VECTOR_ELT(Op, DAG);
case ISD::EXTRACT_VECTOR_ELT: return LowerEXTRACT_VECTOR_ELT(Op, DAG);
case ISD::CONCAT_VECTORS: return LowerCONCAT_VECTORS(Op, DAG);
case ISD::FLT_ROUNDS_: return LowerFLT_ROUNDS_(Op, DAG);
return BB;
}
+/// EmitBasePointerRecalculation - For functions using a base pointer, we
+/// rematerialize it (via the frame pointer).
+void ARMTargetLowering::
+EmitBasePointerRecalculation(MachineInstr *MI, MachineBasicBlock *MBB,
+ MachineBasicBlock *DispatchBB) const {
+ const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+ const ARMBaseInstrInfo *AII = static_cast<const ARMBaseInstrInfo*>(TII);
+ MachineFunction &MF = *MI->getParent()->getParent();
+ ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
+ const ARMBaseRegisterInfo &RI = AII->getRegisterInfo();
+
+ if (!RI.hasBasePointer(MF)) return;
+
+ MachineBasicBlock::iterator MBBI = MI;
+
+ int32_t NumBytes = AFI->getFramePtrSpillOffset();
+ unsigned FramePtr = RI.getFrameRegister(MF);
+ assert(MF.getTarget().getFrameLowering()->hasFP(MF) &&
+ "Base pointer without frame pointer?");
+
+ if (AFI->isThumb2Function())
+ llvm::emitT2RegPlusImmediate(*MBB, MBBI, MI->getDebugLoc(), ARM::R6,
+ FramePtr, -NumBytes, ARMCC::AL, 0, *AII);
+ else if (AFI->isThumbFunction())
+ llvm::emitThumbRegPlusImmediate(*MBB, MBBI, MI->getDebugLoc(), ARM::R6,
+ FramePtr, -NumBytes, *AII, RI);
+ else
+ llvm::emitARMRegPlusImmediate(*MBB, MBBI, MI->getDebugLoc(), ARM::R6,
+ FramePtr, -NumBytes, ARMCC::AL, 0, *AII);
+
+ if (!RI.needsStackRealignment(MF)) return;
+
+ // If there's dynamic realignment, adjust for it.
+ MachineFrameInfo *MFI = MF.getFrameInfo();
+ unsigned MaxAlign = MFI->getMaxAlignment();
+ assert(!AFI->isThumb1OnlyFunction());
+
+ // Emit bic r6, r6, MaxAlign
+ unsigned bicOpc = AFI->isThumbFunction() ? ARM::t2BICri : ARM::BICri;
+ AddDefaultCC(
+ AddDefaultPred(
+ BuildMI(*MBB, MBBI, MI->getDebugLoc(), TII->get(bicOpc), ARM::R6)
+ .addReg(ARM::R6, RegState::Kill)
+ .addImm(MaxAlign - 1)));
+}
+
+/// SetupEntryBlockForSjLj - Insert code into the entry block that creates and
+/// registers the function context.
+void ARMTargetLowering::
+SetupEntryBlockForSjLj(MachineInstr *MI, MachineBasicBlock *MBB,
+ MachineBasicBlock *DispatchBB, int FI) const {
+ const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+ DebugLoc dl = MI->getDebugLoc();
+ MachineFunction *MF = MBB->getParent();
+ MachineRegisterInfo *MRI = &MF->getRegInfo();
+ MachineConstantPool *MCP = MF->getConstantPool();
+ ARMFunctionInfo *AFI = MF->getInfo<ARMFunctionInfo>();
+ const Function *F = MF->getFunction();
+
+ bool isThumb = Subtarget->isThumb();
+ bool isThumb2 = Subtarget->isThumb2();
+
+ unsigned PCLabelId = AFI->createPICLabelUId();
+ unsigned PCAdj = (isThumb || isThumb2) ? 4 : 8;
+ ARMConstantPoolValue *CPV =
+ ARMConstantPoolMBB::Create(F->getContext(), DispatchBB, PCLabelId, PCAdj);
+ unsigned CPI = MCP->getConstantPoolIndex(CPV, 4);
+
+ const TargetRegisterClass *TRC =
+ isThumb ? ARM::tGPRRegisterClass : ARM::GPRRegisterClass;
+
+ // Grab constant pool and fixed stack memory operands.
+ MachineMemOperand *CPMMO =
+ MF->getMachineMemOperand(MachinePointerInfo::getConstantPool(),
+ MachineMemOperand::MOLoad, 4, 4);
+
+ MachineMemOperand *FIMMOSt =
+ MF->getMachineMemOperand(MachinePointerInfo::getFixedStack(FI),
+ MachineMemOperand::MOStore, 4, 4);
+
+ EmitBasePointerRecalculation(MI, MBB, DispatchBB);
+
+ // Load the address of the dispatch MBB into the jump buffer.
+ if (isThumb2) {
+ // Incoming value: jbuf
+ // ldr.n r5, LCPI1_1
+ // orr r5, r5, #1
+ // add r5, pc
+ // str r5, [$jbuf, #+4] ; &jbuf[1]
+ unsigned NewVReg1 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::t2LDRpci), NewVReg1)
+ .addConstantPoolIndex(CPI)
+ .addMemOperand(CPMMO));
+ // Set the low bit because of thumb mode.
+ unsigned NewVReg2 = MRI->createVirtualRegister(TRC);
+ AddDefaultCC(
+ AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::t2ORRri), NewVReg2)
+ .addReg(NewVReg1, RegState::Kill)
+ .addImm(0x01)));
+ unsigned NewVReg3 = MRI->createVirtualRegister(TRC);
+ BuildMI(*MBB, MI, dl, TII->get(ARM::tPICADD), NewVReg3)
+ .addReg(NewVReg2, RegState::Kill)
+ .addImm(PCLabelId);
+ AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::t2STRi12))
+ .addReg(NewVReg3, RegState::Kill)
+ .addFrameIndex(FI)
+ .addImm(36) // &jbuf[1] :: pc
+ .addMemOperand(FIMMOSt));
+ } else if (isThumb) {
+ // Incoming value: jbuf
+ // ldr.n r1, LCPI1_4
+ // add r1, pc
+ // mov r2, #1
+ // orrs r1, r2
+ // add r2, $jbuf, #+4 ; &jbuf[1]
+ // str r1, [r2]
+ unsigned NewVReg1 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::tLDRpci), NewVReg1)
+ .addConstantPoolIndex(CPI)
+ .addMemOperand(CPMMO));
+ unsigned NewVReg2 = MRI->createVirtualRegister(TRC);
+ BuildMI(*MBB, MI, dl, TII->get(ARM::tPICADD), NewVReg2)
+ .addReg(NewVReg1, RegState::Kill)
+ .addImm(PCLabelId);
+ // Set the low bit because of thumb mode.
+ unsigned NewVReg3 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::tMOVi8), NewVReg3)
+ .addReg(ARM::CPSR, RegState::Define)
+ .addImm(1));
+ unsigned NewVReg4 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::tORR), NewVReg4)
+ .addReg(ARM::CPSR, RegState::Define)
+ .addReg(NewVReg2, RegState::Kill)
+ .addReg(NewVReg3, RegState::Kill));
+ unsigned NewVReg5 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::tADDrSPi), NewVReg5)
+ .addFrameIndex(FI)
+ .addImm(36)); // &jbuf[1] :: pc
+ AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::tSTRi))
+ .addReg(NewVReg4, RegState::Kill)
+ .addReg(NewVReg5, RegState::Kill)
+ .addImm(0)
+ .addMemOperand(FIMMOSt));
+ } else {
+ // Incoming value: jbuf
+ // ldr r1, LCPI1_1
+ // add r1, pc, r1
+ // str r1, [$jbuf, #+4] ; &jbuf[1]
+ unsigned NewVReg1 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::LDRi12), NewVReg1)
+ .addConstantPoolIndex(CPI)
+ .addImm(0)
+ .addMemOperand(CPMMO));
+ unsigned NewVReg2 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::PICADD), NewVReg2)
+ .addReg(NewVReg1, RegState::Kill)
+ .addImm(PCLabelId));
+ AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::STRi12))
+ .addReg(NewVReg2, RegState::Kill)
+ .addFrameIndex(FI)
+ .addImm(36) // &jbuf[1] :: pc
+ .addMemOperand(FIMMOSt));
+ }
+}
+
+MachineBasicBlock *ARMTargetLowering::
+EmitSjLjDispatchBlock(MachineInstr *MI, MachineBasicBlock *MBB) const {
+ const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+ DebugLoc dl = MI->getDebugLoc();
+ MachineFunction *MF = MBB->getParent();
+ MachineRegisterInfo *MRI = &MF->getRegInfo();
+ ARMFunctionInfo *AFI = MF->getInfo<ARMFunctionInfo>();
+ MachineFrameInfo *MFI = MF->getFrameInfo();
+ int FI = MFI->getFunctionContextIndex();
+
+ const TargetRegisterClass *TRC =
+ Subtarget->isThumb() ? ARM::tGPRRegisterClass : ARM::GPRRegisterClass;
+
+ // Get a mapping of the call site numbers to all of the landing pads they're
+ // associated with.
+ DenseMap<unsigned, SmallVector<MachineBasicBlock*, 2> > CallSiteNumToLPad;
+ unsigned MaxCSNum = 0;
+ MachineModuleInfo &MMI = MF->getMMI();
+ for (MachineFunction::iterator BB = MF->begin(), E = MF->end(); BB != E; ++BB) {
+ if (!BB->isLandingPad()) continue;
+
+ // FIXME: We should assert that the EH_LABEL is the first MI in the landing
+ // pad.
+ for (MachineBasicBlock::iterator
+ II = BB->begin(), IE = BB->end(); II != IE; ++II) {
+ if (!II->isEHLabel()) continue;
+
+ MCSymbol *Sym = II->getOperand(0).getMCSymbol();
+ if (!MMI.hasCallSiteLandingPad(Sym)) continue;
+
+ SmallVectorImpl<unsigned> &CallSiteIdxs = MMI.getCallSiteLandingPad(Sym);
+ for (SmallVectorImpl<unsigned>::iterator
+ CSI = CallSiteIdxs.begin(), CSE = CallSiteIdxs.end();
+ CSI != CSE; ++CSI) {
+ CallSiteNumToLPad[*CSI].push_back(BB);
+ MaxCSNum = std::max(MaxCSNum, *CSI);
+ }
+ break;
+ }
+ }
+
+ // Get an ordered list of the machine basic blocks for the jump table.
+ std::vector<MachineBasicBlock*> LPadList;
+ SmallPtrSet<MachineBasicBlock*, 64> InvokeBBs;
+ LPadList.reserve(CallSiteNumToLPad.size());
+ for (unsigned I = 1; I <= MaxCSNum; ++I) {
+ SmallVectorImpl<MachineBasicBlock*> &MBBList = CallSiteNumToLPad[I];
+ for (SmallVectorImpl<MachineBasicBlock*>::iterator
+ II = MBBList.begin(), IE = MBBList.end(); II != IE; ++II) {
+ LPadList.push_back(*II);
+ InvokeBBs.insert((*II)->pred_begin(), (*II)->pred_end());
+ }
+ }
+
+ assert(!LPadList.empty() &&
+ "No landing pad destinations for the dispatch jump table!");
+
+ // Create the jump table and associated information.
+ MachineJumpTableInfo *JTI =
+ MF->getOrCreateJumpTableInfo(MachineJumpTableInfo::EK_Inline);
+ unsigned MJTI = JTI->createJumpTableIndex(LPadList);
+ unsigned UId = AFI->createJumpTableUId();
+
+ // Create the MBBs for the dispatch code.
+
+ // Shove the dispatch's address into the return slot in the function context.
+ MachineBasicBlock *DispatchBB = MF->CreateMachineBasicBlock();
+ DispatchBB->setIsLandingPad();
+
+ MachineBasicBlock *TrapBB = MF->CreateMachineBasicBlock();
+ BuildMI(TrapBB, dl, TII->get(Subtarget->isThumb() ? ARM::tTRAP : ARM::TRAP));
+ DispatchBB->addSuccessor(TrapBB);
+
+ MachineBasicBlock *DispContBB = MF->CreateMachineBasicBlock();
+ DispatchBB->addSuccessor(DispContBB);
+
+ // Insert and MBBs.
+ MF->insert(MF->end(), DispatchBB);
+ MF->insert(MF->end(), DispContBB);
+ MF->insert(MF->end(), TrapBB);
+
+ // Insert code into the entry block that creates and registers the function
+ // context.
+ SetupEntryBlockForSjLj(MI, MBB, DispatchBB, FI);
+
+ MachineMemOperand *FIMMOLd =
+ MF->getMachineMemOperand(MachinePointerInfo::getFixedStack(FI),
+ MachineMemOperand::MOLoad |
+ MachineMemOperand::MOVolatile, 4, 4);
+
+ unsigned NumLPads = LPadList.size();
+ if (Subtarget->isThumb2()) {
+ unsigned NewVReg1 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::t2LDRi12), NewVReg1)
+ .addFrameIndex(FI)
+ .addImm(4)
+ .addMemOperand(FIMMOLd));
+
+ if (NumLPads < 256) {
+ AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::t2CMPri))
+ .addReg(NewVReg1)
+ .addImm(LPadList.size()));
+ } else {
+ unsigned VReg1 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::t2MOVi16), VReg1)
+ .addImm(NumLPads & 0xFFFF));
+
+ unsigned VReg2 = VReg1;
+ if ((NumLPads & 0xFFFF0000) != 0) {
+ VReg2 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::t2MOVTi16), VReg2)
+ .addReg(VReg1)
+ .addImm(NumLPads >> 16));
+ }
+
+ AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::t2CMPrr))
+ .addReg(NewVReg1)
+ .addReg(VReg2));
+ }
+
+ BuildMI(DispatchBB, dl, TII->get(ARM::t2Bcc))
+ .addMBB(TrapBB)
+ .addImm(ARMCC::HI)
+ .addReg(ARM::CPSR);
+
+ unsigned NewVReg3 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::t2LEApcrelJT),NewVReg3)
+ .addJumpTableIndex(MJTI)
+ .addImm(UId));
+
+ unsigned NewVReg4 = MRI->createVirtualRegister(TRC);
+ AddDefaultCC(
+ AddDefaultPred(
+ BuildMI(DispContBB, dl, TII->get(ARM::t2ADDrs), NewVReg4)
+ .addReg(NewVReg3, RegState::Kill)
+ .addReg(NewVReg1)
+ .addImm(ARM_AM::getSORegOpc(ARM_AM::lsl, 2))));
+
+ BuildMI(DispContBB, dl, TII->get(ARM::t2BR_JT))
+ .addReg(NewVReg4, RegState::Kill)
+ .addReg(NewVReg1)
+ .addJumpTableIndex(MJTI)
+ .addImm(UId);
+ } else if (Subtarget->isThumb()) {
+ unsigned NewVReg1 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::tLDRspi), NewVReg1)
+ .addFrameIndex(FI)
+ .addImm(1)
+ .addMemOperand(FIMMOLd));
+
+ if (NumLPads < 256) {
+ AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::tCMPi8))
+ .addReg(NewVReg1)
+ .addImm(NumLPads));
+ } else {
+ MachineConstantPool *ConstantPool = MF->getConstantPool();
+ Type *Int32Ty = Type::getInt32Ty(MF->getFunction()->getContext());
+ const Constant *C = ConstantInt::get(Int32Ty, NumLPads);
+
+ // MachineConstantPool wants an explicit alignment.
+ unsigned Align = getTargetData()->getPrefTypeAlignment(Int32Ty);
+ if (Align == 0)
+ Align = getTargetData()->getTypeAllocSize(C->getType());
+ unsigned Idx = ConstantPool->getConstantPoolIndex(C, Align);
+
+ unsigned VReg1 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::tLDRpci))
+ .addReg(VReg1, RegState::Define)
+ .addConstantPoolIndex(Idx));
+ AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::tCMPr))
+ .addReg(NewVReg1)
+ .addReg(VReg1));
+ }
+
+ BuildMI(DispatchBB, dl, TII->get(ARM::tBcc))
+ .addMBB(TrapBB)
+ .addImm(ARMCC::HI)
+ .addReg(ARM::CPSR);
+
+ unsigned NewVReg2 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tLSLri), NewVReg2)
+ .addReg(ARM::CPSR, RegState::Define)
+ .addReg(NewVReg1)
+ .addImm(2));
+
+ unsigned NewVReg3 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tLEApcrelJT), NewVReg3)
+ .addJumpTableIndex(MJTI)
+ .addImm(UId));
+
+ unsigned NewVReg4 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tADDrr), NewVReg4)
+ .addReg(ARM::CPSR, RegState::Define)
+ .addReg(NewVReg2, RegState::Kill)
+ .addReg(NewVReg3));
+
+ MachineMemOperand *JTMMOLd =
+ MF->getMachineMemOperand(MachinePointerInfo::getJumpTable(),
+ MachineMemOperand::MOLoad, 4, 4);
+
+ unsigned NewVReg5 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tLDRi), NewVReg5)
+ .addReg(NewVReg4, RegState::Kill)
+ .addImm(0)
+ .addMemOperand(JTMMOLd));
+
+ unsigned NewVReg6 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tADDrr), NewVReg6)
+ .addReg(ARM::CPSR, RegState::Define)
+ .addReg(NewVReg5, RegState::Kill)
+ .addReg(NewVReg3));
+
+ BuildMI(DispContBB, dl, TII->get(ARM::tBR_JTr))
+ .addReg(NewVReg6, RegState::Kill)
+ .addJumpTableIndex(MJTI)
+ .addImm(UId);
+ } else {
+ unsigned NewVReg1 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::LDRi12), NewVReg1)
+ .addFrameIndex(FI)
+ .addImm(4)
+ .addMemOperand(FIMMOLd));
+
+ if (NumLPads < 256) {
+ AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::CMPri))
+ .addReg(NewVReg1)
+ .addImm(NumLPads));
+ } else if (Subtarget->hasV6T2Ops() && isUInt<16>(NumLPads)) {
+ unsigned VReg1 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::MOVi16), VReg1)
+ .addImm(NumLPads & 0xFFFF));
+
+ unsigned VReg2 = VReg1;
+ if ((NumLPads & 0xFFFF0000) != 0) {
+ VReg2 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::MOVTi16), VReg2)
+ .addReg(VReg1)
+ .addImm(NumLPads >> 16));
+ }
+
+ AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::CMPrr))
+ .addReg(NewVReg1)
+ .addReg(VReg2));
+ } else {
+ MachineConstantPool *ConstantPool = MF->getConstantPool();
+ Type *Int32Ty = Type::getInt32Ty(MF->getFunction()->getContext());
+ const Constant *C = ConstantInt::get(Int32Ty, NumLPads);
+
+ // MachineConstantPool wants an explicit alignment.
+ unsigned Align = getTargetData()->getPrefTypeAlignment(Int32Ty);
+ if (Align == 0)
+ Align = getTargetData()->getTypeAllocSize(C->getType());
+ unsigned Idx = ConstantPool->getConstantPoolIndex(C, Align);
+
+ unsigned VReg1 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::LDRcp))
+ .addReg(VReg1, RegState::Define)
+ .addConstantPoolIndex(Idx)
+ .addImm(0));
+ AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::CMPrr))
+ .addReg(NewVReg1)
+ .addReg(VReg1, RegState::Kill));
+ }
+
+ BuildMI(DispatchBB, dl, TII->get(ARM::Bcc))
+ .addMBB(TrapBB)
+ .addImm(ARMCC::HI)
+ .addReg(ARM::CPSR);
+
+ unsigned NewVReg3 = MRI->createVirtualRegister(TRC);
+ AddDefaultCC(
+ AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::MOVsi), NewVReg3)
+ .addReg(NewVReg1)
+ .addImm(ARM_AM::getSORegOpc(ARM_AM::lsl, 2))));
+ unsigned NewVReg4 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::LEApcrelJT), NewVReg4)
+ .addJumpTableIndex(MJTI)
+ .addImm(UId));
+
+ MachineMemOperand *JTMMOLd =
+ MF->getMachineMemOperand(MachinePointerInfo::getJumpTable(),
+ MachineMemOperand::MOLoad, 4, 4);
+ unsigned NewVReg5 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(
+ BuildMI(DispContBB, dl, TII->get(ARM::LDRrs), NewVReg5)
+ .addReg(NewVReg3, RegState::Kill)
+ .addReg(NewVReg4)
+ .addImm(0)
+ .addMemOperand(JTMMOLd));
+
+ BuildMI(DispContBB, dl, TII->get(ARM::BR_JTadd))
+ .addReg(NewVReg5, RegState::Kill)
+ .addReg(NewVReg4)
+ .addJumpTableIndex(MJTI)
+ .addImm(UId);
+ }
+
+ // Add the jump table entries as successors to the MBB.
+ MachineBasicBlock *PrevMBB = 0;
+ for (std::vector<MachineBasicBlock*>::iterator
+ I = LPadList.begin(), E = LPadList.end(); I != E; ++I) {
+ MachineBasicBlock *CurMBB = *I;
+ if (PrevMBB != CurMBB)
+ DispContBB->addSuccessor(CurMBB);
+ PrevMBB = CurMBB;
+ }
+
+ // N.B. the order the invoke BBs are processed in doesn't matter here.
+ const ARMBaseInstrInfo *AII = static_cast<const ARMBaseInstrInfo*>(TII);
+ const ARMBaseRegisterInfo &RI = AII->getRegisterInfo();
+ const unsigned *SavedRegs = RI.getCalleeSavedRegs(MF);
+ SmallVector<MachineBasicBlock*, 64> MBBLPads;
+ for (SmallPtrSet<MachineBasicBlock*, 64>::iterator
+ I = InvokeBBs.begin(), E = InvokeBBs.end(); I != E; ++I) {
+ MachineBasicBlock *BB = *I;
+
+ // Remove the landing pad successor from the invoke block and replace it
+ // with the new dispatch block.
+ SmallVector<MachineBasicBlock*, 4> Successors(BB->succ_begin(),
+ BB->succ_end());
+ while (!Successors.empty()) {
+ MachineBasicBlock *SMBB = Successors.pop_back_val();
+ if (SMBB->isLandingPad()) {
+ BB->removeSuccessor(SMBB);
+ MBBLPads.push_back(SMBB);
+ }
+ }
+
+ BB->addSuccessor(DispatchBB);
+
+ // Find the invoke call and mark all of the callee-saved registers as
+ // 'implicit defined' so that they're spilled. This prevents code from
+ // moving instructions to before the EH block, where they will never be
+ // executed.
+ for (MachineBasicBlock::reverse_iterator
+ II = BB->rbegin(), IE = BB->rend(); II != IE; ++II) {
+ if (!II->getDesc().isCall()) continue;
+
+ DenseMap<unsigned, bool> DefRegs;
+ for (MachineInstr::mop_iterator
+ OI = II->operands_begin(), OE = II->operands_end();
+ OI != OE; ++OI) {
+ if (!OI->isReg()) continue;
+ DefRegs[OI->getReg()] = true;
+ }
+
+ MachineInstrBuilder MIB(&*II);
+
+ for (unsigned i = 0; SavedRegs[i] != 0; ++i) {
+ unsigned Reg = SavedRegs[i];
+ if (Subtarget->isThumb2() &&
+ !ARM::tGPRRegisterClass->contains(Reg) &&
+ !ARM::hGPRRegisterClass->contains(Reg))
+ continue;
+ else if (Subtarget->isThumb1Only() &&
+ !ARM::tGPRRegisterClass->contains(Reg))
+ continue;
+ else if (!Subtarget->isThumb() &&
+ !ARM::GPRRegisterClass->contains(Reg))
+ continue;
+ if (!DefRegs[Reg])
+ MIB.addReg(Reg, RegState::ImplicitDefine | RegState::Dead);
+ }
+
+ break;
+ }
+ }
+
+ // Mark all former landing pads as non-landing pads. The dispatch is the only
+ // landing pad now.
+ for (SmallVectorImpl<MachineBasicBlock*>::iterator
+ I = MBBLPads.begin(), E = MBBLPads.end(); I != E; ++I)
+ (*I)->setIsLandingPad(false);
+
+ // The instruction is gone now.
+ MI->eraseFromParent();
+
+ return MBB;
+}
+
static
MachineBasicBlock *OtherSucc(MachineBasicBlock *MBB, MachineBasicBlock *Succ) {
for (MachineBasicBlock::succ_iterator I = MBB->succ_begin(),
MI->eraseFromParent(); // The pseudo instruction is gone now.
return BB;
}
+
+ case ARM::Int_eh_sjlj_setjmp:
+ case ARM::Int_eh_sjlj_setjmp_nofp:
+ case ARM::tInt_eh_sjlj_setjmp:
+ case ARM::t2Int_eh_sjlj_setjmp:
+ case ARM::t2Int_eh_sjlj_setjmp_nofp:
+ EmitSjLjDispatchBlock(MI, BB);
+ return BB;
+
+ case ARM::ABS:
+ case ARM::t2ABS: {
+ // To insert an ABS instruction, we have to insert the
+ // diamond control-flow pattern. The incoming instruction knows the
+ // source vreg to test against 0, the destination vreg to set,
+ // the condition code register to branch on, the
+ // true/false values to select between, and a branch opcode to use.
+ // It transforms
+ // V1 = ABS V0
+ // into
+ // V2 = MOVS V0
+ // BCC (branch to SinkBB if V0 >= 0)
+ // RSBBB: V3 = RSBri V2, 0 (compute ABS if V2 < 0)
+ // SinkBB: V1 = PHI(V2, V3)
+ const BasicBlock *LLVM_BB = BB->getBasicBlock();
+ MachineFunction::iterator BBI = BB;
+ ++BBI;
+ MachineFunction *Fn = BB->getParent();
+ MachineBasicBlock *RSBBB = Fn->CreateMachineBasicBlock(LLVM_BB);
+ MachineBasicBlock *SinkBB = Fn->CreateMachineBasicBlock(LLVM_BB);
+ Fn->insert(BBI, RSBBB);
+ Fn->insert(BBI, SinkBB);
+
+ unsigned int ABSSrcReg = MI->getOperand(1).getReg();
+ unsigned int ABSDstReg = MI->getOperand(0).getReg();
+ bool isThumb2 = Subtarget->isThumb2();
+ MachineRegisterInfo &MRI = Fn->getRegInfo();
+ // In Thumb mode S must not be specified if source register is the SP or
+ // PC and if destination register is the SP, so restrict register class
+ unsigned NewMovDstReg = MRI.createVirtualRegister(
+ isThumb2 ? ARM::rGPRRegisterClass : ARM::GPRRegisterClass);
+ unsigned NewRsbDstReg = MRI.createVirtualRegister(
+ isThumb2 ? ARM::rGPRRegisterClass : ARM::GPRRegisterClass);
+
+ // Transfer the remainder of BB and its successor edges to sinkMBB.
+ SinkBB->splice(SinkBB->begin(), BB,
+ llvm::next(MachineBasicBlock::iterator(MI)),
+ BB->end());
+ SinkBB->transferSuccessorsAndUpdatePHIs(BB);
+
+ BB->addSuccessor(RSBBB);
+ BB->addSuccessor(SinkBB);
+
+ // fall through to SinkMBB
+ RSBBB->addSuccessor(SinkBB);
+
+ // insert a movs at the end of BB
+ BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2MOVr : ARM::MOVr),
+ NewMovDstReg)
+ .addReg(ABSSrcReg, RegState::Kill)
+ .addImm((unsigned)ARMCC::AL).addReg(0)
+ .addReg(ARM::CPSR, RegState::Define);
+
+ // insert a bcc with opposite CC to ARMCC::MI at the end of BB
+ BuildMI(BB, dl,
+ TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc)).addMBB(SinkBB)
+ .addImm(ARMCC::getOppositeCondition(ARMCC::MI)).addReg(ARM::CPSR);
+
+ // insert rsbri in RSBBB
+ // Note: BCC and rsbri will be converted into predicated rsbmi
+ // by if-conversion pass
+ BuildMI(*RSBBB, RSBBB->begin(), dl,
+ TII->get(isThumb2 ? ARM::t2RSBri : ARM::RSBri), NewRsbDstReg)
+ .addReg(NewMovDstReg, RegState::Kill)
+ .addImm(0).addImm((unsigned)ARMCC::AL).addReg(0).addReg(0);
+
+ // insert PHI in SinkBB,
+ // reuse ABSDstReg to not change uses of ABS instruction
+ BuildMI(*SinkBB, SinkBB->begin(), dl,
+ TII->get(ARM::PHI), ABSDstReg)
+ .addReg(NewRsbDstReg).addMBB(RSBBB)
+ .addReg(NewMovDstReg).addMBB(BB);
+
+ // remove ABS instruction
+ MI->eraseFromParent();
+
+ // return last added BB
+ return SinkBB;
+ }
}
}
void ARMTargetLowering::AdjustInstrPostInstrSelection(MachineInstr *MI,
SDNode *Node) const {
- const MCInstrDesc &MCID = MI->getDesc();
- if (!MCID.hasPostISelHook()) {
+ const MCInstrDesc *MCID = &MI->getDesc();
+ if (!MCID->hasPostISelHook()) {
assert(!convertAddSubFlagsOpcode(MI->getOpcode()) &&
"Pseudo flag-setting opcodes must be marked with 'hasPostISelHook'");
return;
// operand is still set to noreg. If needed, set the optional operand's
// register to CPSR, and remove the redundant implicit def.
//
- // e.g. ADCS (...opt:%noreg, CPSR<imp-def>) -> ADC (... opt:CPSR<def>).
+ // e.g. ADCS (..., CPSR<imp-def>) -> ADC (... opt:CPSR<def>).
// Rename pseudo opcodes.
unsigned NewOpc = convertAddSubFlagsOpcode(MI->getOpcode());
if (NewOpc) {
const ARMBaseInstrInfo *TII =
static_cast<const ARMBaseInstrInfo*>(getTargetMachine().getInstrInfo());
- MI->setDesc(TII->get(NewOpc));
+ MCID = &TII->get(NewOpc);
+
+ assert(MCID->getNumOperands() == MI->getDesc().getNumOperands() + 1 &&
+ "converted opcode should be the same except for cc_out");
+
+ MI->setDesc(*MCID);
+
+ // Add the optional cc_out operand
+ MI->addOperand(MachineOperand::CreateReg(0, /*isDef=*/true));
}
- unsigned ccOutIdx = MCID.getNumOperands() - 1;
+ unsigned ccOutIdx = MCID->getNumOperands() - 1;
// Any ARM instruction that sets the 's' bit should specify an optional
// "cc_out" operand in the last operand position.
- if (!MCID.hasOptionalDef() || !MCID.OpInfo[ccOutIdx].isOptionalDef()) {
+ if (!MCID->hasOptionalDef() || !MCID->OpInfo[ccOutIdx].isOptionalDef()) {
assert(!NewOpc && "Optional cc_out operand required");
return;
}
// since we already have an optional CPSR def.
bool definesCPSR = false;
bool deadCPSR = false;
- for (unsigned i = MCID.getNumOperands(), e = MI->getNumOperands();
+ for (unsigned i = MCID->getNumOperands(), e = MI->getNumOperands();
i != e; ++i) {
const MachineOperand &MO = MI->getOperand(i);
if (MO.isReg() && MO.isDef() && MO.getReg() == ARM::CPSR) {
SDValue BasePtr = LD->getBasePtr();
SDValue NewLD1 = DAG.getLoad(MVT::i32, DL, LD->getChain(), BasePtr,
LD->getPointerInfo(), LD->isVolatile(),
- LD->isNonTemporal(), LD->getAlignment());
+ LD->isNonTemporal(), LD->isInvariant(),
+ LD->getAlignment());
SDValue OffsetPtr = DAG.getNode(ISD::ADD, DL, MVT::i32, BasePtr,
DAG.getConstant(4, MVT::i32));
SDValue NewLD2 = DAG.getLoad(MVT::i32, DL, NewLD1.getValue(1), OffsetPtr,
LD->getPointerInfo(), LD->isVolatile(),
- LD->isNonTemporal(),
+ LD->isNonTemporal(), LD->isInvariant(),
std::min(4U, LD->getAlignment() / 2));
DAG.ReplaceAllUsesOfValueWith(SDValue(LD, 1), NewLD2.getValue(1));
}
}
+static bool memOpAlign(unsigned DstAlign, unsigned SrcAlign,
+ unsigned AlignCheck) {
+ return ((SrcAlign == 0 || SrcAlign % AlignCheck == 0) &&
+ (DstAlign == 0 || DstAlign % AlignCheck == 0));
+}
+
+EVT ARMTargetLowering::getOptimalMemOpType(uint64_t Size,
+ unsigned DstAlign, unsigned SrcAlign,
+ bool IsZeroVal,
+ bool MemcpyStrSrc,
+ MachineFunction &MF) const {
+ const Function *F = MF.getFunction();
+
+ // See if we can use NEON instructions for this...
+ if (IsZeroVal &&
+ !F->hasFnAttr(Attribute::NoImplicitFloat) &&
+ Subtarget->hasNEON()) {
+ if (memOpAlign(SrcAlign, DstAlign, 16) && Size >= 16) {
+ return MVT::v4i32;
+ } else if (memOpAlign(SrcAlign, DstAlign, 8) && Size >= 8) {
+ return MVT::v2i32;
+ }
+ }
+
+ // Lowering to i32/i16 if the size permits.
+ if (Size >= 4) {
+ return MVT::i32;
+ } else if (Size >= 2) {
+ return MVT::i16;
+ }
+
+ // Let the target-independent logic figure it out.
+ return MVT::Other;
+}
+
static bool isLegalT1AddressImmediate(int64_t V, EVT VT) {
if (V < 0)
return false;