setTargetDAGCombine(ISD::FP_TO_SINT);
setTargetDAGCombine(ISD::FP_TO_UINT);
setTargetDAGCombine(ISD::FDIV);
+
+ setLoadExtAction(ISD::EXTLOAD, MVT::v4i8, Expand);
}
computeRegisterProperties();
}
// 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)
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
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::
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
// 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)
+ II = MBBList.begin(), IE = MBBList.end(); II != IE; ++II) {
LPadList.push_back(*II);
+ InvokeBBs.insert((*II)->pred_begin(), (*II)->pred_end());
+ }
}
assert(!LPadList.empty() &&
// Shove the dispatch's address into the return slot in the function context.
MachineBasicBlock *DispatchBB = MF->CreateMachineBasicBlock();
DispatchBB->setIsLandingPad();
- MBB->addSuccessor(DispatchBB);
MachineBasicBlock *TrapBB = MF->CreateMachineBasicBlock();
BuildMI(TrapBB, dl, TII->get(Subtarget->isThumb() ? ARM::tTRAP : ARM::TRAP));
MachineBasicBlock *DispContBB = MF->CreateMachineBasicBlock();
DispatchBB->addSuccessor(DispContBB);
- // Insert and renumber MBBs.
- MachineBasicBlock *Last = &MF->back();
+ // Insert and MBBs.
MF->insert(MF->end(), DispatchBB);
MF->insert(MF->end(), DispContBB);
MF->insert(MF->end(), TrapBB);
- MF->RenumberBlocks(Last);
// Insert code into the entry block that creates and registers the function
// context.
.addFrameIndex(FI)
.addImm(1)
.addMemOperand(FIMMOLd));
+
AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::tCMPi8))
.addReg(NewVReg1)
.addImm(LPadList.size()));
}
// 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)
- DispContBB->addSuccessor(*I);
+ 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 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.
+ // 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)
+ // SinkBB: V1 = PHI(V2, V3)
const BasicBlock *LLVM_BB = BB->getBasicBlock();
MachineFunction::iterator BBI = BB;
++BBI;
.addReg(ARM::CPSR, RegState::Define);
// insert a bcc with opposite CC to ARMCC::MI at the end of BB
- BuildMI(BB, dl,
+ 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,
+ 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,
+ // insert PHI in SinkBB,
// reuse ABSDstReg to not change uses of ABS instruction
BuildMI(*SinkBB, SinkBB->begin(), dl,
TII->get(ARM::PHI), ABSDstReg)
.addReg(NewMovDstReg).addMBB(BB);
// remove ABS instruction
- MI->eraseFromParent();
+ MI->eraseFromParent();
// return last added BB
return SinkBB;