#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::VSETCC, VT.getSimpleVT(), Custom);
+ setOperationAction(ISD::SETCC, VT.getSimpleVT(), Custom);
setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT.getSimpleVT(), Custom);
if (ElemTy != MVT::i32) {
setOperationAction(ISD::SINT_TO_FP, VT.getSimpleVT(), Expand);
RegInfo = TM.getRegisterInfo();
Itins = TM.getInstrItineraryData();
+ setBooleanVectorContents(ZeroOrNegativeOneBooleanContent);
+
if (Subtarget->isTargetDarwin()) {
// Uses VFP for Thumb libfuncs if available.
if (Subtarget->isThumb() && Subtarget->hasVFP2()) {
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
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::SETCC, MVT::v2f64, Expand);
setOperationAction(ISD::FNEG, MVT::v2f64, Expand);
setOperationAction(ISD::FABS, MVT::v2f64, Expand);
setOperationAction(ISD::FSQRT, MVT::v2f64, Expand);
setOperationAction(ISD::SDIV, MVT::v8i8, Custom);
setOperationAction(ISD::UDIV, MVT::v4i16, Custom);
setOperationAction(ISD::UDIV, MVT::v8i8, Custom);
- setOperationAction(ISD::VSETCC, MVT::v1i64, Expand);
- setOperationAction(ISD::VSETCC, MVT::v2i64, Expand);
+ setOperationAction(ISD::SETCC, MVT::v1i64, Expand);
+ setOperationAction(ISD::SETCC, MVT::v2i64, Expand);
// Neon does not have single instruction SINT_TO_FP and UINT_TO_FP with
// a destination type that is wider than the source.
setOperationAction(ISD::SINT_TO_FP, MVT::v4i16, Custom);
setTargetDAGCombine(ISD::FP_TO_SINT);
setTargetDAGCombine(ISD::FP_TO_UINT);
setTargetDAGCombine(ISD::FDIV);
+
+ setLoadExtAction(ISD::EXTLOAD, MVT::v4i8, Expand);
}
computeRegisterProperties();
setOperationAction(ISD::ATOMIC_LOAD_OR, MVT::i64, Custom);
setOperationAction(ISD::ATOMIC_LOAD_XOR, MVT::i64, Custom);
setOperationAction(ISD::ATOMIC_SWAP, MVT::i64, Custom);
+ setOperationAction(ISD::ATOMIC_CMP_SWAP, MVT::i64, Custom);
// Automatically insert fences (dmb ist) around ATOMIC_SWAP etc.
setInsertFencesForAtomic(true);
} else {
setOperationAction(ISD::ATOMIC_LOAD_MAX, MVT::i32, Expand);
setOperationAction(ISD::ATOMIC_LOAD_UMIN, MVT::i32, Expand);
setOperationAction(ISD::ATOMIC_LOAD_UMAX, MVT::i32, Expand);
- setOperationAction(ISD::ATOMIC_LOAD, MVT::i32, Expand);
- setOperationAction(ISD::ATOMIC_STORE, MVT::i32, Expand);
+ // Mark ATOMIC_LOAD and ATOMIC_STORE custom so we can handle the
+ // Unordered/Monotonic case.
+ setOperationAction(ISD::ATOMIC_LOAD, MVT::i32, Custom);
+ setOperationAction(ISD::ATOMIC_STORE, MVT::i32, Custom);
// Since the libcalls include locking, fold in the fences
setShouldFoldAtomicFences(true);
}
}
}
+EVT ARMTargetLowering::getSetCCResultType(EVT VT) const {
+ if (!VT.isVector()) return getPointerTy();
+ return VT.changeVectorElementTypeToInteger();
+}
+
/// getRegClassFor - Return the register class that should be used for the
/// specified value type.
TargetRegisterClass *ARMTargetLowering::getRegClassFor(EVT VT) const {
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 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=*/false,
+ /*AlwaysInline=*/true,
MachinePointerInfo(0),
MachinePointerInfo(0)));
const GlobalValue *GV = G->getGlobal();
// Create a constant pool entry for the callee address
unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
- ARMConstantPoolValue *CPV = new ARMConstantPoolValue(GV,
- ARMPCLabelIndex,
- ARMCP::CPValue, 0);
+ ARMConstantPoolValue *CPV =
+ ARMConstantPoolConstant::Create(GV, ARMPCLabelIndex, ARMCP::CPValue, 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);
// 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);
// tBX takes a register source operand.
if (isARMFunc && Subtarget->isThumb1Only() && !Subtarget->hasV5TOps()) {
unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
- ARMConstantPoolValue *CPV = new ARMConstantPoolValue(GV,
- ARMPCLabelIndex,
- ARMCP::CPValue, 4);
+ ARMConstantPoolValue *CPV =
+ ARMConstantPoolConstant::Create(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,
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,
} else {
unsigned PCAdj = Subtarget->isThumb() ? 4 : 8;
ARMPCLabelIndex = AFI->createPICLabelUId();
- ARMConstantPoolValue *CPV = new ARMConstantPoolValue(BA, ARMPCLabelIndex,
- ARMCP::CPBlockAddress,
- PCAdj);
+ ARMConstantPoolValue *CPV =
+ ARMConstantPoolConstant::Create(BA, ARMPCLabelIndex,
+ ARMCP::CPBlockAddress, PCAdj);
CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4);
}
CPAddr = DAG.getNode(ARMISD::Wrapper, DL, PtrVT, CPAddr);
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
ARMConstantPoolValue *CPV =
- new ARMConstantPoolValue(GA->getGlobal(), ARMPCLabelIndex,
- ARMCP::CPValue, PCAdj, ARMCP::TLSGD, true);
+ ARMConstantPoolConstant::Create(GA->getGlobal(), ARMPCLabelIndex,
+ ARMCP::CPValue, PCAdj, ARMCP::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,
// Initial exec model.
unsigned char PCAdj = Subtarget->isThumb() ? 4 : 8;
ARMConstantPoolValue *CPV =
- new ARMConstantPoolValue(GA->getGlobal(), ARMPCLabelIndex,
- ARMCP::CPValue, PCAdj, ARMCP::GOTTPOFF, true);
+ ARMConstantPoolConstant::Create(GA->getGlobal(), ARMPCLabelIndex,
+ ARMCP::CPValue, PCAdj, ARMCP::GOTTPOFF,
+ true);
Offset = DAG.getTargetConstantPool(CPV, PtrVT, 4);
Offset = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, Offset);
Offset = DAG.getLoad(PtrVT, dl, Chain, Offset,
false, false, 0);
} else {
// local exec model
- ARMConstantPoolValue *CPV = new ARMConstantPoolValue(GV, ARMCP::TPOFF);
+ ARMConstantPoolValue *CPV =
+ ARMConstantPoolConstant::Create(GV, ARMCP::TPOFF);
Offset = DAG.getTargetConstantPool(CPV, PtrVT, 4);
Offset = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, Offset);
Offset = DAG.getLoad(PtrVT, dl, Chain, Offset,
if (RelocM == Reloc::PIC_) {
bool UseGOTOFF = GV->hasLocalLinkage() || GV->hasHiddenVisibility();
ARMConstantPoolValue *CPV =
- new ARMConstantPoolValue(GV, UseGOTOFF ? ARMCP::GOTOFF : ARMCP::GOT);
+ ARMConstantPoolConstant::Create(GV,
+ UseGOTOFF ? ARMCP::GOTOFF : ARMCP::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(),
ARMPCLabelIndex = AFI->createPICLabelUId();
unsigned PCAdj = (RelocM != Reloc::PIC_) ? 0 : (Subtarget->isThumb()?4:8);
ARMConstantPoolValue *CPV =
- new ARMConstantPoolValue(GV, ARMPCLabelIndex, ARMCP::CPValue, PCAdj);
+ ARMConstantPoolConstant::Create(GV, ARMPCLabelIndex, ARMCP::CPValue,
+ PCAdj);
CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4);
}
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
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,
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);
}
unsigned PCAdj = (RelocM != Reloc::PIC_)
? 0 : (Subtarget->isThumb() ? 4 : 8);
ARMConstantPoolValue *CPV =
- new ARMConstantPoolValue(MF.getFunction(), ARMPCLabelIndex,
- ARMCP::CPLSDA, PCAdj);
+ ARMConstantPoolConstant::Create(MF.getFunction(), ARMPCLabelIndex,
+ ARMCP::CPLSDA, PCAdj);
CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4);
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
SDValue Result =
SDValue ARMcc;
SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
SDValue Cmp = getARMCmp(LHS, RHS, CC, ARMcc, DAG, dl);
- return DAG.getNode(ARMISD::CMOV, dl, VT, FalseVal, TrueVal, ARMcc, CCR, Cmp);
+ return DAG.getNode(ARMISD::CMOV, dl, VT, FalseVal, TrueVal, ARMcc, CCR,Cmp);
}
ARMCC::CondCodes CondCode, CondCode2;
}
// 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...
unsigned HalfSize = EltSize / 2;
if (isSigned) {
int64_t SExtVal = C->getSExtValue();
- if ((SExtVal >> HalfSize) != (SExtVal >> EltSize))
+ if (SExtVal != SExtVal << (64 - HalfSize) >> (64 - HalfSize))
return false;
} else {
if ((C->getZExtValue() >> HalfSize) != 0)
Op.getOperand(1), Op.getOperand(2));
}
+static SDValue LowerAtomicLoadStore(SDValue Op, SelectionDAG &DAG) {
+ // Monotonic load/store is legal for all targets
+ if (cast<AtomicSDNode>(Op)->getOrdering() <= Monotonic)
+ return Op;
+
+ // Aquire/Release load/store is not legal for targets without a
+ // dmb or equivalent available.
+ return SDValue();
+}
+
+
static void
-ReplaceATOMIC_BINARY_64(SDNode *Node, SmallVectorImpl<SDValue>& Results,
- SelectionDAG &DAG, unsigned NewOp) {
- EVT T = Node->getValueType(0);
+ReplaceATOMIC_OP_64(SDNode *Node, SmallVectorImpl<SDValue>& Results,
+ SelectionDAG &DAG, unsigned NewOp) {
DebugLoc dl = Node->getDebugLoc();
- assert (T == MVT::i64 && "Only know how to expand i64 atomics");
-
- SDValue Chain = Node->getOperand(0);
- SDValue In1 = Node->getOperand(1);
- SDValue In2L = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32,
- Node->getOperand(2), DAG.getIntPtrConstant(0));
- SDValue In2H = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32,
- Node->getOperand(2), DAG.getIntPtrConstant(1));
- SDValue Ops[] = { Chain, In1, In2L, In2H };
+ assert (Node->getValueType(0) == MVT::i64 &&
+ "Only know how to expand i64 atomics");
+
+ SmallVector<SDValue, 6> Ops;
+ Ops.push_back(Node->getOperand(0)); // Chain
+ Ops.push_back(Node->getOperand(1)); // Ptr
+ // Low part of Val1
+ Ops.push_back(DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32,
+ Node->getOperand(2), DAG.getIntPtrConstant(0)));
+ // High part of Val1
+ Ops.push_back(DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32,
+ Node->getOperand(2), DAG.getIntPtrConstant(1)));
+ if (NewOp == ARMISD::ATOMCMPXCHG64_DAG) {
+ // High part of Val1
+ Ops.push_back(DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32,
+ Node->getOperand(3), DAG.getIntPtrConstant(0)));
+ // High part of Val2
+ Ops.push_back(DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32,
+ Node->getOperand(3), DAG.getIntPtrConstant(1)));
+ }
SDVTList Tys = DAG.getVTList(MVT::i32, MVT::i32, MVT::Other);
SDValue Result =
- DAG.getMemIntrinsicNode(NewOp, dl, Tys, Ops, 4, MVT::i64,
+ DAG.getMemIntrinsicNode(NewOp, dl, Tys, Ops.data(), Ops.size(), MVT::i64,
cast<MemSDNode>(Node)->getMemOperand());
- SDValue OpsF[] = { Result.getValue(0), Result.getValue(1)};
+ SDValue OpsF[] = { Result.getValue(0), Result.getValue(1) };
Results.push_back(DAG.getNode(ISD::BUILD_PAIR, dl, MVT::i64, OpsF, 2));
Results.push_back(Result.getValue(2));
}
case ISD::SRL_PARTS:
case ISD::SRA_PARTS: return LowerShiftRightParts(Op, DAG);
case ISD::CTTZ: return LowerCTTZ(Op.getNode(), DAG, Subtarget);
- case ISD::VSETCC: return LowerVSETCC(Op, DAG);
+ 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::EXTRACT_VECTOR_ELT: return LowerEXTRACT_VECTOR_ELT(Op, DAG);
case ISD::ADDE:
case ISD::SUBC:
case ISD::SUBE: return LowerADDC_ADDE_SUBC_SUBE(Op, DAG);
+ case ISD::ATOMIC_LOAD:
+ case ISD::ATOMIC_STORE: return LowerAtomicLoadStore(Op, DAG);
}
return SDValue();
}
Res = Expand64BitShift(N, DAG, Subtarget);
break;
case ISD::ATOMIC_LOAD_ADD:
- ReplaceATOMIC_BINARY_64(N, Results, DAG, ARMISD::ATOMADD64_DAG);
+ ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMADD64_DAG);
return;
case ISD::ATOMIC_LOAD_AND:
- ReplaceATOMIC_BINARY_64(N, Results, DAG, ARMISD::ATOMAND64_DAG);
+ ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMAND64_DAG);
return;
case ISD::ATOMIC_LOAD_NAND:
- ReplaceATOMIC_BINARY_64(N, Results, DAG, ARMISD::ATOMNAND64_DAG);
+ ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMNAND64_DAG);
return;
case ISD::ATOMIC_LOAD_OR:
- ReplaceATOMIC_BINARY_64(N, Results, DAG, ARMISD::ATOMOR64_DAG);
+ ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMOR64_DAG);
return;
case ISD::ATOMIC_LOAD_SUB:
- ReplaceATOMIC_BINARY_64(N, Results, DAG, ARMISD::ATOMSUB64_DAG);
+ ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMSUB64_DAG);
return;
case ISD::ATOMIC_LOAD_XOR:
- ReplaceATOMIC_BINARY_64(N, Results, DAG, ARMISD::ATOMXOR64_DAG);
+ ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMXOR64_DAG);
return;
case ISD::ATOMIC_SWAP:
- ReplaceATOMIC_BINARY_64(N, Results, DAG, ARMISD::ATOMSWAP64_DAG);
+ ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMSWAP64_DAG);
+ return;
+ case ISD::ATOMIC_CMP_SWAP:
+ ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMCMPXCHG64_DAG);
return;
- //case ISD::ATOMIC_CMP_SWAP:
- // ReplaceATOMIC_CMPXCHG_64(N, Results, DAG);
}
if (Res.getNode())
Results.push_back(Res);
// cmp dest, oldval
// bne exitMBB
BB = loop1MBB;
- AddDefaultPred(BuildMI(BB, dl, TII->get(ldrOpc), dest).addReg(ptr));
+ MachineInstrBuilder MIB = BuildMI(BB, dl, TII->get(ldrOpc), dest).addReg(ptr);
+ if (ldrOpc == ARM::t2LDREX)
+ MIB.addImm(0);
+ AddDefaultPred(MIB);
AddDefaultPred(BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2CMPrr : ARM::CMPrr))
.addReg(dest).addReg(oldval));
BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc))
// cmp scratch, #0
// bne loop1MBB
BB = loop2MBB;
- AddDefaultPred(BuildMI(BB, dl, TII->get(strOpc), scratch).addReg(newval)
- .addReg(ptr));
+ MIB = BuildMI(BB, dl, TII->get(strOpc), scratch).addReg(newval).addReg(ptr);
+ if (strOpc == ARM::t2STREX)
+ MIB.addImm(0);
+ AddDefaultPred(MIB);
AddDefaultPred(BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2CMPri : ARM::CMPri))
.addReg(scratch).addImm(0));
BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc))
// bne- loopMBB
// fallthrough --> exitMBB
BB = loopMBB;
- AddDefaultPred(BuildMI(BB, dl, TII->get(ldrOpc), dest).addReg(ptr));
+ MachineInstrBuilder MIB = BuildMI(BB, dl, TII->get(ldrOpc), dest).addReg(ptr);
+ if (ldrOpc == ARM::t2LDREX)
+ MIB.addImm(0);
+ AddDefaultPred(MIB);
if (BinOpcode) {
// operand order needs to go the other way for NAND
if (BinOpcode == ARM::BICrr || BinOpcode == ARM::t2BICrr)
addReg(dest).addReg(incr)).addReg(0);
}
- AddDefaultPred(BuildMI(BB, dl, TII->get(strOpc), scratch).addReg(scratch2)
- .addReg(ptr));
+ MIB = BuildMI(BB, dl, TII->get(strOpc), scratch).addReg(scratch2).addReg(ptr);
+ if (strOpc == ARM::t2STREX)
+ MIB.addImm(0);
+ AddDefaultPred(MIB);
AddDefaultPred(BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2CMPri : ARM::CMPri))
.addReg(scratch).addImm(0));
BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc))
// bne- loopMBB
// fallthrough --> exitMBB
BB = loopMBB;
- AddDefaultPred(BuildMI(BB, dl, TII->get(ldrOpc), dest).addReg(ptr));
+ MachineInstrBuilder MIB = BuildMI(BB, dl, TII->get(ldrOpc), dest).addReg(ptr);
+ if (ldrOpc == ARM::t2LDREX)
+ MIB.addImm(0);
+ AddDefaultPred(MIB);
// Sign extend the value, if necessary.
if (signExtend && extendOpc) {
BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2MOVCCr : ARM::MOVCCr), scratch2)
.addReg(oldval).addReg(incr).addImm(Cond).addReg(ARM::CPSR);
- AddDefaultPred(BuildMI(BB, dl, TII->get(strOpc), scratch).addReg(scratch2)
- .addReg(ptr));
+ MIB = BuildMI(BB, dl, TII->get(strOpc), scratch).addReg(scratch2).addReg(ptr);
+ if (strOpc == ARM::t2STREX)
+ MIB.addImm(0);
+ AddDefaultPred(MIB);
AddDefaultPred(BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2CMPri : ARM::CMPri))
.addReg(scratch).addImm(0));
BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc))
MachineBasicBlock *
ARMTargetLowering::EmitAtomicBinary64(MachineInstr *MI, MachineBasicBlock *BB,
unsigned Op1, unsigned Op2,
- bool NeedsCarry) const {
+ bool NeedsCarry, bool IsCmpxchg) const {
// This also handles ATOMIC_SWAP, indicated by Op1==0.
const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
unsigned strOpc = isThumb2 ? ARM::t2STREXD : ARM::STREXD;
MachineBasicBlock *loopMBB = MF->CreateMachineBasicBlock(LLVM_BB);
+ MachineBasicBlock *contBB = 0, *cont2BB = 0;
+ if (IsCmpxchg) {
+ contBB = MF->CreateMachineBasicBlock(LLVM_BB);
+ cont2BB = MF->CreateMachineBasicBlock(LLVM_BB);
+ }
MachineBasicBlock *exitMBB = MF->CreateMachineBasicBlock(LLVM_BB);
MF->insert(It, loopMBB);
+ if (IsCmpxchg) {
+ MF->insert(It, contBB);
+ MF->insert(It, cont2BB);
+ }
MF->insert(It, exitMBB);
// Transfer the remainder of BB and its successor edges to exitMBB.
// Note that the registers are explicitly specified because there is not any
// way to force the register allocator to allocate a register pair.
//
- // FIXME: The hardcoded registers are not necessary for Thumb2, but we
+ // FIXME: The hardcoded registers are not necessary for Thumb2, but we
// need to properly enforce the restriction that the two output registers
// for ldrexd must be different.
BB = loopMBB;
// Copy r2/r3 into dest. (This copy will normally be coalesced.)
BuildMI(BB, dl, TII->get(TargetOpcode::COPY), destlo).addReg(ARM::R2);
BuildMI(BB, dl, TII->get(TargetOpcode::COPY), desthi).addReg(ARM::R3);
- if (Op1) {
+
+ if (IsCmpxchg) {
+ // Add early exit
+ for (unsigned i = 0; i < 2; i++) {
+ AddDefaultPred(BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2CMPrr :
+ ARM::CMPrr))
+ .addReg(i == 0 ? destlo : desthi)
+ .addReg(i == 0 ? vallo : valhi));
+ BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc))
+ .addMBB(exitMBB).addImm(ARMCC::NE).addReg(ARM::CPSR);
+ BB->addSuccessor(exitMBB);
+ BB->addSuccessor(i == 0 ? contBB : cont2BB);
+ BB = (i == 0 ? contBB : cont2BB);
+ }
+
+ // Copy to physregs for strexd
+ unsigned setlo = MI->getOperand(5).getReg();
+ unsigned sethi = MI->getOperand(6).getReg();
+ BuildMI(BB, dl, TII->get(TargetOpcode::COPY), ARM::R0).addReg(setlo);
+ BuildMI(BB, dl, TII->get(TargetOpcode::COPY), ARM::R1).addReg(sethi);
+ } else if (Op1) {
// Perform binary operation
AddDefaultPred(BuildMI(BB, dl, TII->get(Op1), ARM::R0)
.addReg(destlo).addReg(vallo))
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);
+
+ if (Subtarget->isThumb2()) {
+ unsigned NewVReg1 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::t2LDRi12), NewVReg1)
+ .addFrameIndex(FI)
+ .addImm(4)
+ .addMemOperand(FIMMOLd));
+ AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::t2CMPri))
+ .addReg(NewVReg1)
+ .addImm(LPadList.size()));
+ BuildMI(DispatchBB, dl, TII->get(ARM::t2Bcc))
+ .addMBB(TrapBB)
+ .addImm(ARMCC::HI)
+ .addReg(ARM::CPSR);
+
+ unsigned NewVReg2 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::t2LEApcrelJT),NewVReg2)
+ .addJumpTableIndex(MJTI)
+ .addImm(UId));
+
+ unsigned NewVReg3 = MRI->createVirtualRegister(TRC);
+ AddDefaultCC(
+ AddDefaultPred(
+ BuildMI(DispContBB, dl, TII->get(ARM::t2ADDrs), NewVReg3)
+ .addReg(NewVReg2, RegState::Kill)
+ .addReg(NewVReg1)
+ .addImm(ARM_AM::getSORegOpc(ARM_AM::lsl, 2))));
+
+ BuildMI(DispContBB, dl, TII->get(ARM::t2BR_JT))
+ .addReg(NewVReg3, 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));
+
+ AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::tCMPi8))
+ .addReg(NewVReg1)
+ .addImm(LPadList.size()));
+ 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));
+ AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::CMPri))
+ .addReg(NewVReg1)
+ .addImm(LPadList.size()));
+ BuildMI(DispatchBB, dl, TII->get(ARM::Bcc))
+ .addMBB(TrapBB)
+ .addImm(ARMCC::HI)
+ .addReg(ARM::CPSR);
+
+ unsigned NewVReg2 = MRI->createVirtualRegister(TRC);
+ AddDefaultCC(
+ AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::MOVsi), NewVReg2)
+ .addReg(NewVReg1)
+ .addImm(ARM_AM::getSORegOpc(ARM_AM::lsl, 2))));
+ unsigned NewVReg3 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::LEApcrelJT), NewVReg3)
+ .addJumpTableIndex(MJTI)
+ .addImm(UId));
+
+ MachineMemOperand *JTMMOLd =
+ MF->getMachineMemOperand(MachinePointerInfo::getJumpTable(),
+ MachineMemOperand::MOLoad, 4, 4);
+ unsigned NewVReg4 = MRI->createVirtualRegister(TRC);
+ AddDefaultPred(
+ BuildMI(DispContBB, dl, TII->get(ARM::LDRrs), NewVReg4)
+ .addReg(NewVReg2, RegState::Kill)
+ .addReg(NewVReg3)
+ .addImm(0)
+ .addMemOperand(JTMMOLd));
+
+ BuildMI(DispContBB, dl, TII->get(ARM::BR_JTadd))
+ .addReg(NewVReg4, RegState::Kill)
+ .addReg(NewVReg3)
+ .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.
+ for (MachineBasicBlock::succ_iterator
+ SI = BB->succ_begin(), SE = BB->succ_end(); SI != SE; ++SI) {
+ MachineBasicBlock *SMBB = *SI;
+ 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) {
+ if (!TRC->contains(SavedRegs[i])) continue;
+ if (!DefRegs[SavedRegs[i]])
+ MIB.addReg(SavedRegs[i], 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->dump();
llvm_unreachable("Unexpected instr type to insert");
}
+ // The Thumb2 pre-indexed stores have the same MI operands, they just
+ // define them differently in the .td files from the isel patterns, so
+ // they need pseudos.
+ case ARM::t2STR_preidx:
+ MI->setDesc(TII->get(ARM::t2STR_PRE));
+ return BB;
+ case ARM::t2STRB_preidx:
+ MI->setDesc(TII->get(ARM::t2STRB_PRE));
+ return BB;
+ case ARM::t2STRH_preidx:
+ MI->setDesc(TII->get(ARM::t2STRH_PRE));
+ return BB;
+
case ARM::STRi_preidx:
case ARM::STRBi_preidx: {
unsigned NewOpc = MI->getOpcode() == ARM::STRi_preidx ?
case ARM::ATOMADD6432:
return EmitAtomicBinary64(MI, BB, isThumb2 ? ARM::t2ADDrr : ARM::ADDrr,
- isThumb2 ? ARM::t2ADCrr : ARM::ADCrr, true);
+ isThumb2 ? ARM::t2ADCrr : ARM::ADCrr,
+ /*NeedsCarry*/ true);
case ARM::ATOMSUB6432:
return EmitAtomicBinary64(MI, BB, isThumb2 ? ARM::t2SUBrr : ARM::SUBrr,
- isThumb2 ? ARM::t2SBCrr : ARM::SBCrr, true);
+ isThumb2 ? ARM::t2SBCrr : ARM::SBCrr,
+ /*NeedsCarry*/ true);
case ARM::ATOMOR6432:
return EmitAtomicBinary64(MI, BB, isThumb2 ? ARM::t2ORRrr : ARM::ORRrr,
- isThumb2 ? ARM::t2ORRrr : ARM::ORRrr, false);
+ isThumb2 ? ARM::t2ORRrr : ARM::ORRrr);
case ARM::ATOMXOR6432:
return EmitAtomicBinary64(MI, BB, isThumb2 ? ARM::t2EORrr : ARM::EORrr,
- isThumb2 ? ARM::t2EORrr : ARM::EORrr, false);
+ isThumb2 ? ARM::t2EORrr : ARM::EORrr);
case ARM::ATOMAND6432:
return EmitAtomicBinary64(MI, BB, isThumb2 ? ARM::t2ANDrr : ARM::ANDrr,
- isThumb2 ? ARM::t2ANDrr : ARM::ANDrr, false);
+ isThumb2 ? ARM::t2ANDrr : ARM::ANDrr);
case ARM::ATOMSWAP6432:
return EmitAtomicBinary64(MI, BB, 0, 0, false);
+ case ARM::ATOMCMPXCHG6432:
+ return EmitAtomicBinary64(MI, BB, isThumb2 ? ARM::t2SUBrr : ARM::SUBrr,
+ isThumb2 ? ARM::t2SBCrr : ARM::SBCrr,
+ /*NeedsCarry*/ false, /*IsCmpxchg*/true);
case ARM::tMOVCCr_pseudo: {
// To "insert" a SELECT_CC instruction, we actually have to insert the
BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc))
.addMBB(destMBB).addImm(ARMCC::EQ).addReg(ARM::CPSR);
- BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2B : ARM::B))
- .addMBB(exitMBB);
+ if (isThumb2)
+ AddDefaultPred(BuildMI(BB, dl, TII->get(ARM::t2B)).addMBB(exitMBB));
+ else
+ BuildMI(BB, dl, TII->get(ARM::B)) .addMBB(exitMBB);
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 {
- // Adjust potentially 's' setting instructions after isel, i.e. ADC, SBC,
- // RSB, RSC. Coming out of isel, they have an implicit CPSR def, but the
- // optional operand is not filled in. If the carry bit is used, then change
- // the optional operand to CPSR. Otherwise, remove the CPSR implicit def.
const MCInstrDesc &MCID = MI->getDesc();
- if (Node->hasAnyUseOfValue(1)) {
- MachineOperand &MO = MI->getOperand(MCID.getNumOperands() - 2);
- MO.setReg(ARM::CPSR);
- MO.setIsDef(true);
- } else {
- 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) {
- MI->RemoveOperand(i);
- break;
- }
+ if (!MCID.hasPostISelHook()) {
+ assert(!convertAddSubFlagsOpcode(MI->getOpcode()) &&
+ "Pseudo flag-setting opcodes must be marked with 'hasPostISelHook'");
+ return;
+ }
+
+ // Adjust potentially 's' setting instructions after isel, i.e. ADC, SBC, RSB,
+ // RSC. Coming out of isel, they have an implicit CPSR def, but the optional
+ // 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>).
+
+ // Rename pseudo opcodes.
+ unsigned NewOpc = convertAddSubFlagsOpcode(MI->getOpcode());
+ if (NewOpc) {
+ const ARMBaseInstrInfo *TII =
+ static_cast<const ARMBaseInstrInfo*>(getTargetMachine().getInstrInfo());
+ MI->setDesc(TII->get(NewOpc));
+ }
+ 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()) {
+ assert(!NewOpc && "Optional cc_out operand required");
+ return;
+ }
+ // Look for an implicit def of CPSR added by MachineInstr ctor. Remove it
+ // since we already have an optional CPSR def.
+ bool definesCPSR = false;
+ bool deadCPSR = false;
+ 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) {
+ definesCPSR = true;
+ if (MO.isDead())
+ deadCPSR = true;
+ MI->RemoveOperand(i);
+ break;
}
}
+ if (!definesCPSR) {
+ assert(!NewOpc && "Optional cc_out operand required");
+ return;
+ }
+ assert(deadCPSR == !Node->hasAnyUseOfValue(1) && "inconsistent dead flag");
+ if (deadCPSR) {
+ assert(!MI->getOperand(ccOutIdx).getReg() &&
+ "expect uninitialized optional cc_out operand");
+ return;
+ }
+
+ // If this instruction was defined with an optional CPSR def and its dag node
+ // had a live implicit CPSR def, then activate the optional CPSR def.
+ MachineOperand &MO = MI->getOperand(ccOutIdx);
+ MO.setReg(ARM::CPSR);
+ MO.setIsDef(true);
}
//===----------------------------------------------------------------------===//
SDValue FalseVal = N->getOperand(0);
SDValue TrueVal = N->getOperand(1);
SDValue ARMcc = N->getOperand(2);
- ARMCC::CondCodes CC = (ARMCC::CondCodes)cast<ConstantSDNode>(ARMcc)->getZExtValue();
+ ARMCC::CondCodes CC =
+ (ARMCC::CondCodes)cast<ConstantSDNode>(ARMcc)->getZExtValue();
// Simplify
// mov r1, r0
// movne r0, y
/// FIXME: Turn this into a target neutral optimization?
SDValue Res;
- if (CC == ARMCC::NE && FalseVal == RHS) {
+ if (CC == ARMCC::NE && FalseVal == RHS && FalseVal != LHS) {
Res = DAG.getNode(ARMISD::CMOV, dl, VT, LHS, TrueVal, ARMcc,
N->getOperand(3), Cmp);
} else if (CC == ARMCC::EQ && TrueVal == RHS) {
return RCPair(0U, ARM::GPRRegisterClass);
case 'h': // High regs or no regs.
if (Subtarget->isThumb())
- return RCPair(0U, ARM::hGPRRegisterClass);
+ return RCPair(0U, ARM::hGPRRegisterClass);
break;
case 'r':
return RCPair(0U, ARM::GPRRegisterClass);
break;
case 'x':
if (VT == MVT::f32)
- return RCPair(0U, ARM::SPR_8RegisterClass);
+ return RCPair(0U, ARM::SPR_8RegisterClass);
if (VT.getSizeInBits() == 64)
- return RCPair(0U, ARM::DPR_8RegisterClass);
+ return RCPair(0U, ARM::DPR_8RegisterClass);
if (VT.getSizeInBits() == 128)
- return RCPair(0U, ARM::QPR_8RegisterClass);
+ return RCPair(0U, ARM::QPR_8RegisterClass);
break;
case 't':
if (VT == MVT::f32)
- return RCPair(0U, ARM::SPRRegisterClass);
+ return RCPair(0U, ARM::SPRRegisterClass);
break;
}
}
switch (ConstraintLetter) {
case 'j':
- // Constant suitable for movw, must be between 0 and
- // 65535.
- if (Subtarget->hasV6T2Ops())
- if (CVal >= 0 && CVal <= 65535)
- break;
- return;
+ // Constant suitable for movw, must be between 0 and
+ // 65535.
+ if (Subtarget->hasV6T2Ops())
+ if (CVal >= 0 && CVal <= 65535)
+ break;
+ return;
case 'I':
if (Subtarget->isThumb1Only()) {
// This must be a constant between 0 and 255, for ADD
return false;
}
-int ARM::getVFPf32Imm(const APFloat &FPImm) {
- APInt Imm = FPImm.bitcastToAPInt();
- uint32_t Sign = Imm.lshr(31).getZExtValue() & 1;
- int32_t Exp = (Imm.lshr(23).getSExtValue() & 0xff) - 127; // -126 to 127
- int64_t Mantissa = Imm.getZExtValue() & 0x7fffff; // 23 bits
-
- // We can handle 4 bits of mantissa.
- // mantissa = (16+UInt(e:f:g:h))/16.
- if (Mantissa & 0x7ffff)
- return -1;
- Mantissa >>= 19;
- if ((Mantissa & 0xf) != Mantissa)
- return -1;
-
- // We can handle 3 bits of exponent: exp == UInt(NOT(b):c:d)-3
- if (Exp < -3 || Exp > 4)
- return -1;
- Exp = ((Exp+3) & 0x7) ^ 4;
-
- return ((int)Sign << 7) | (Exp << 4) | Mantissa;
-}
-
-int ARM::getVFPf64Imm(const APFloat &FPImm) {
- APInt Imm = FPImm.bitcastToAPInt();
- uint64_t Sign = Imm.lshr(63).getZExtValue() & 1;
- int64_t Exp = (Imm.lshr(52).getSExtValue() & 0x7ff) - 1023; // -1022 to 1023
- uint64_t Mantissa = Imm.getZExtValue() & 0xfffffffffffffLL;
-
- // We can handle 4 bits of mantissa.
- // mantissa = (16+UInt(e:f:g:h))/16.
- if (Mantissa & 0xffffffffffffLL)
- return -1;
- Mantissa >>= 48;
- if ((Mantissa & 0xf) != Mantissa)
- return -1;
-
- // We can handle 3 bits of exponent: exp == UInt(NOT(b):c:d)-3
- if (Exp < -3 || Exp > 4)
- return -1;
- Exp = ((Exp+3) & 0x7) ^ 4;
-
- return ((int)Sign << 7) | (Exp << 4) | Mantissa;
-}
-
bool ARM::isBitFieldInvertedMask(unsigned v) {
if (v == 0xffffffff)
return 0;
if (!Subtarget->hasVFP3())
return false;
if (VT == MVT::f32)
- return ARM::getVFPf32Imm(Imm) != -1;
+ return ARM_AM::getFP32Imm(Imm) != -1;
if (VT == MVT::f64)
- return ARM::getVFPf64Imm(Imm) != -1;
+ return ARM_AM::getFP64Imm(Imm) != -1;
return false;
}