X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FCodeGen%2FSelectionDAG%2FSelectionDAGISel.cpp;h=7213a5b9318c2ecc24dc2b3144865e8f74c7c926;hb=1f7a90d7936a9a6278365ea8b0521c7ff17432c3;hp=68b9146adfe1ad96d41f8d34ef4905dd41ae4957;hpb=30e6740f2e3d518ce1cfcd484ef728ac5764a645;p=oota-llvm.git diff --git a/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp b/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp index 68b9146adfe..7213a5b9318 100644 --- a/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp +++ b/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp @@ -11,24 +11,18 @@ // //===----------------------------------------------------------------------===// -#define DEBUG_TYPE "isel" +#include "llvm/CodeGen/GCStrategy.h" #include "ScheduleDAGSDNodes.h" #include "SelectionDAGBuilder.h" -#include "llvm/CodeGen/FunctionLoweringInfo.h" -#include "llvm/CodeGen/SelectionDAGISel.h" +#include "llvm/ADT/PostOrderIterator.h" +#include "llvm/ADT/Statistic.h" #include "llvm/Analysis/AliasAnalysis.h" #include "llvm/Analysis/BranchProbabilityInfo.h" -#include "llvm/Analysis/DebugInfo.h" -#include "llvm/Constants.h" -#include "llvm/Function.h" -#include "llvm/InlineAsm.h" -#include "llvm/Instructions.h" -#include "llvm/Intrinsics.h" -#include "llvm/IntrinsicInst.h" -#include "llvm/LLVMContext.h" -#include "llvm/Module.h" +#include "llvm/Analysis/CFG.h" +#include "llvm/Analysis/TargetLibraryInfo.h" +#include "llvm/CodeGen/Analysis.h" #include "llvm/CodeGen/FastISel.h" -#include "llvm/CodeGen/GCStrategy.h" +#include "llvm/CodeGen/FunctionLoweringInfo.h" #include "llvm/CodeGen/GCMetadata.h" #include "llvm/CodeGen/MachineFrameInfo.h" #include "llvm/CodeGen/MachineFunction.h" @@ -38,36 +32,150 @@ #include "llvm/CodeGen/ScheduleHazardRecognizer.h" #include "llvm/CodeGen/SchedulerRegistry.h" #include "llvm/CodeGen/SelectionDAG.h" -#include "llvm/Target/TargetRegisterInfo.h" -#include "llvm/Target/TargetIntrinsicInfo.h" -#include "llvm/Target/TargetInstrInfo.h" -#include "llvm/Target/TargetLowering.h" -#include "llvm/Target/TargetMachine.h" -#include "llvm/Target/TargetOptions.h" -#include "llvm/Transforms/Utils/BasicBlockUtils.h" +#include "llvm/CodeGen/SelectionDAGISel.h" +#include "llvm/CodeGen/WinEHFuncInfo.h" +#include "llvm/IR/Constants.h" +#include "llvm/IR/DebugInfo.h" +#include "llvm/IR/Function.h" +#include "llvm/IR/InlineAsm.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/IntrinsicInst.h" +#include "llvm/IR/Intrinsics.h" +#include "llvm/IR/LLVMContext.h" +#include "llvm/IR/Module.h" +#include "llvm/MC/MCAsmInfo.h" #include "llvm/Support/Compiler.h" #include "llvm/Support/Debug.h" #include "llvm/Support/ErrorHandling.h" #include "llvm/Support/Timer.h" #include "llvm/Support/raw_ostream.h" -#include "llvm/ADT/PostOrderIterator.h" -#include "llvm/ADT/Statistic.h" +#include "llvm/Target/TargetInstrInfo.h" +#include "llvm/Target/TargetIntrinsicInfo.h" +#include "llvm/Target/TargetLowering.h" +#include "llvm/Target/TargetMachine.h" +#include "llvm/Target/TargetOptions.h" +#include "llvm/Target/TargetRegisterInfo.h" +#include "llvm/Target/TargetSubtargetInfo.h" +#include "llvm/Transforms/Utils/BasicBlockUtils.h" #include using namespace llvm; +#define DEBUG_TYPE "isel" + STATISTIC(NumFastIselFailures, "Number of instructions fast isel failed on"); STATISTIC(NumFastIselSuccess, "Number of instructions fast isel selected"); STATISTIC(NumFastIselBlocks, "Number of blocks selected entirely by fast isel"); STATISTIC(NumDAGBlocks, "Number of blocks selected using DAG"); STATISTIC(NumDAGIselRetries,"Number of times dag isel has to try another path"); +STATISTIC(NumEntryBlocks, "Number of entry blocks encountered"); +STATISTIC(NumFastIselFailLowerArguments, + "Number of entry blocks where fast isel failed to lower arguments"); + +#ifndef NDEBUG +static cl::opt +EnableFastISelVerbose2("fast-isel-verbose2", cl::Hidden, + cl::desc("Enable extra verbose messages in the \"fast\" " + "instruction selector")); + + // Terminators +STATISTIC(NumFastIselFailRet,"Fast isel fails on Ret"); +STATISTIC(NumFastIselFailBr,"Fast isel fails on Br"); +STATISTIC(NumFastIselFailSwitch,"Fast isel fails on Switch"); +STATISTIC(NumFastIselFailIndirectBr,"Fast isel fails on IndirectBr"); +STATISTIC(NumFastIselFailInvoke,"Fast isel fails on Invoke"); +STATISTIC(NumFastIselFailResume,"Fast isel fails on Resume"); +STATISTIC(NumFastIselFailUnreachable,"Fast isel fails on Unreachable"); + + // Standard binary operators... +STATISTIC(NumFastIselFailAdd,"Fast isel fails on Add"); +STATISTIC(NumFastIselFailFAdd,"Fast isel fails on FAdd"); +STATISTIC(NumFastIselFailSub,"Fast isel fails on Sub"); +STATISTIC(NumFastIselFailFSub,"Fast isel fails on FSub"); +STATISTIC(NumFastIselFailMul,"Fast isel fails on Mul"); +STATISTIC(NumFastIselFailFMul,"Fast isel fails on FMul"); +STATISTIC(NumFastIselFailUDiv,"Fast isel fails on UDiv"); +STATISTIC(NumFastIselFailSDiv,"Fast isel fails on SDiv"); +STATISTIC(NumFastIselFailFDiv,"Fast isel fails on FDiv"); +STATISTIC(NumFastIselFailURem,"Fast isel fails on URem"); +STATISTIC(NumFastIselFailSRem,"Fast isel fails on SRem"); +STATISTIC(NumFastIselFailFRem,"Fast isel fails on FRem"); + + // Logical operators... +STATISTIC(NumFastIselFailAnd,"Fast isel fails on And"); +STATISTIC(NumFastIselFailOr,"Fast isel fails on Or"); +STATISTIC(NumFastIselFailXor,"Fast isel fails on Xor"); + + // Memory instructions... +STATISTIC(NumFastIselFailAlloca,"Fast isel fails on Alloca"); +STATISTIC(NumFastIselFailLoad,"Fast isel fails on Load"); +STATISTIC(NumFastIselFailStore,"Fast isel fails on Store"); +STATISTIC(NumFastIselFailAtomicCmpXchg,"Fast isel fails on AtomicCmpXchg"); +STATISTIC(NumFastIselFailAtomicRMW,"Fast isel fails on AtomicRWM"); +STATISTIC(NumFastIselFailFence,"Fast isel fails on Frence"); +STATISTIC(NumFastIselFailGetElementPtr,"Fast isel fails on GetElementPtr"); + + // Convert instructions... +STATISTIC(NumFastIselFailTrunc,"Fast isel fails on Trunc"); +STATISTIC(NumFastIselFailZExt,"Fast isel fails on ZExt"); +STATISTIC(NumFastIselFailSExt,"Fast isel fails on SExt"); +STATISTIC(NumFastIselFailFPTrunc,"Fast isel fails on FPTrunc"); +STATISTIC(NumFastIselFailFPExt,"Fast isel fails on FPExt"); +STATISTIC(NumFastIselFailFPToUI,"Fast isel fails on FPToUI"); +STATISTIC(NumFastIselFailFPToSI,"Fast isel fails on FPToSI"); +STATISTIC(NumFastIselFailUIToFP,"Fast isel fails on UIToFP"); +STATISTIC(NumFastIselFailSIToFP,"Fast isel fails on SIToFP"); +STATISTIC(NumFastIselFailIntToPtr,"Fast isel fails on IntToPtr"); +STATISTIC(NumFastIselFailPtrToInt,"Fast isel fails on PtrToInt"); +STATISTIC(NumFastIselFailBitCast,"Fast isel fails on BitCast"); + + // Other instructions... +STATISTIC(NumFastIselFailICmp,"Fast isel fails on ICmp"); +STATISTIC(NumFastIselFailFCmp,"Fast isel fails on FCmp"); +STATISTIC(NumFastIselFailPHI,"Fast isel fails on PHI"); +STATISTIC(NumFastIselFailSelect,"Fast isel fails on Select"); +STATISTIC(NumFastIselFailCall,"Fast isel fails on Call"); +STATISTIC(NumFastIselFailShl,"Fast isel fails on Shl"); +STATISTIC(NumFastIselFailLShr,"Fast isel fails on LShr"); +STATISTIC(NumFastIselFailAShr,"Fast isel fails on AShr"); +STATISTIC(NumFastIselFailVAArg,"Fast isel fails on VAArg"); +STATISTIC(NumFastIselFailExtractElement,"Fast isel fails on ExtractElement"); +STATISTIC(NumFastIselFailInsertElement,"Fast isel fails on InsertElement"); +STATISTIC(NumFastIselFailShuffleVector,"Fast isel fails on ShuffleVector"); +STATISTIC(NumFastIselFailExtractValue,"Fast isel fails on ExtractValue"); +STATISTIC(NumFastIselFailInsertValue,"Fast isel fails on InsertValue"); +STATISTIC(NumFastIselFailLandingPad,"Fast isel fails on LandingPad"); + +// Intrinsic instructions... +STATISTIC(NumFastIselFailIntrinsicCall, "Fast isel fails on Intrinsic call"); +STATISTIC(NumFastIselFailSAddWithOverflow, + "Fast isel fails on sadd.with.overflow"); +STATISTIC(NumFastIselFailUAddWithOverflow, + "Fast isel fails on uadd.with.overflow"); +STATISTIC(NumFastIselFailSSubWithOverflow, + "Fast isel fails on ssub.with.overflow"); +STATISTIC(NumFastIselFailUSubWithOverflow, + "Fast isel fails on usub.with.overflow"); +STATISTIC(NumFastIselFailSMulWithOverflow, + "Fast isel fails on smul.with.overflow"); +STATISTIC(NumFastIselFailUMulWithOverflow, + "Fast isel fails on umul.with.overflow"); +STATISTIC(NumFastIselFailFrameaddress, "Fast isel fails on Frameaddress"); +STATISTIC(NumFastIselFailSqrt, "Fast isel fails on sqrt call"); +STATISTIC(NumFastIselFailStackMap, "Fast isel fails on StackMap call"); +STATISTIC(NumFastIselFailPatchPoint, "Fast isel fails on PatchPoint call"); +#endif static cl::opt EnableFastISelVerbose("fast-isel-verbose", cl::Hidden, cl::desc("Enable verbose messages in the \"fast\" " "instruction selector")); -static cl::opt -EnableFastISelAbort("fast-isel-abort", cl::Hidden, - cl::desc("Enable abort calls when \"fast\" instruction fails")); +static cl::opt EnableFastISelAbort( + "fast-isel-abort", cl::Hidden, + cl::desc("Enable abort calls when \"fast\" instruction selection " + "fails to lower an instruction: 0 disable the abort, 1 will " + "abort but for args, calls and terminators, 2 will also " + "abort for argument lowering, and 3 will never fallback " + "to SelectionDAG.")); static cl::opt UseMBPI("use-mbpi", @@ -75,6 +183,10 @@ UseMBPI("use-mbpi", cl::init(true), cl::Hidden); #ifndef NDEBUG +static cl::opt +FilterDAGBasicBlockName("filter-view-dags", cl::Hidden, + cl::desc("Only display the basic block whose name " + "matches this for all view-*-dags options")); static cl::opt ViewDAGCombine1("view-dag-combine1-dags", cl::Hidden, cl::desc("Pop up a window to show dags before the first " @@ -126,7 +238,7 @@ MachinePassRegistry RegisterScheduler::Registry; static cl::opt > ISHeuristic("pre-RA-sched", - cl::init(&createDefaultScheduler), + cl::init(&createDefaultScheduler), cl::Hidden, cl::desc("Instruction schedulers available (before register" " allocation):")); @@ -135,22 +247,63 @@ defaultListDAGScheduler("default", "Best scheduler for the target", createDefaultScheduler); namespace llvm { + //===--------------------------------------------------------------------===// + /// \brief This class is used by SelectionDAGISel to temporarily override + /// the optimization level on a per-function basis. + class OptLevelChanger { + SelectionDAGISel &IS; + CodeGenOpt::Level SavedOptLevel; + bool SavedFastISel; + + public: + OptLevelChanger(SelectionDAGISel &ISel, + CodeGenOpt::Level NewOptLevel) : IS(ISel) { + SavedOptLevel = IS.OptLevel; + if (NewOptLevel == SavedOptLevel) + return; + IS.OptLevel = NewOptLevel; + IS.TM.setOptLevel(NewOptLevel); + SavedFastISel = IS.TM.Options.EnableFastISel; + if (NewOptLevel == CodeGenOpt::None) + IS.TM.setFastISel(true); + DEBUG(dbgs() << "\nChanging optimization level for Function " + << IS.MF->getFunction()->getName() << "\n"); + DEBUG(dbgs() << "\tBefore: -O" << SavedOptLevel + << " ; After: -O" << NewOptLevel << "\n"); + } + + ~OptLevelChanger() { + if (IS.OptLevel == SavedOptLevel) + return; + DEBUG(dbgs() << "\nRestoring optimization level for Function " + << IS.MF->getFunction()->getName() << "\n"); + DEBUG(dbgs() << "\tBefore: -O" << IS.OptLevel + << " ; After: -O" << SavedOptLevel << "\n"); + IS.OptLevel = SavedOptLevel; + IS.TM.setOptLevel(SavedOptLevel); + IS.TM.setFastISel(SavedFastISel); + } + }; + //===--------------------------------------------------------------------===// /// createDefaultScheduler - This creates an instruction scheduler appropriate /// for the target. ScheduleDAGSDNodes* createDefaultScheduler(SelectionDAGISel *IS, CodeGenOpt::Level OptLevel) { - const TargetLowering &TLI = IS->getTargetLowering(); + const TargetLowering *TLI = IS->TLI; + const TargetSubtargetInfo &ST = IS->MF->getSubtarget(); - if (OptLevel == CodeGenOpt::None) + if (OptLevel == CodeGenOpt::None || + (ST.enableMachineScheduler() && ST.enableMachineSchedDefaultSched()) || + TLI->getSchedulingPreference() == Sched::Source) return createSourceListDAGScheduler(IS, OptLevel); - if (TLI.getSchedulingPreference() == Sched::Latency) - return createTDListDAGScheduler(IS, OptLevel); - if (TLI.getSchedulingPreference() == Sched::RegPressure) + if (TLI->getSchedulingPreference() == Sched::RegPressure) return createBURRListDAGScheduler(IS, OptLevel); - if (TLI.getSchedulingPreference() == Sched::Hybrid) + if (TLI->getSchedulingPreference() == Sched::Hybrid) return createHybridListDAGScheduler(IS, OptLevel); - assert(TLI.getSchedulingPreference() == Sched::ILP && + if (TLI->getSchedulingPreference() == Sched::VLIW) + return createVLIWDAGScheduler(IS, OptLevel); + assert(TLI->getSchedulingPreference() == Sched::ILP && "Unknown sched type!"); return createILPListDAGScheduler(IS, OptLevel); } @@ -173,13 +326,12 @@ TargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI, "'usesCustomInserter', it must implement " "TargetLowering::EmitInstrWithCustomInserter!"; #endif - llvm_unreachable(0); - return 0; + llvm_unreachable(nullptr); } void TargetLowering::AdjustInstrPostInstrSelection(MachineInstr *MI, SDNode *Node) const { - assert(!MI->getDesc().hasPostISelHook() && + assert(!MI->hasPostISelHook() && "If a target marks an instruction with 'hasPostISelHook', " "it must implement TargetLowering::AdjustInstrPostInstrSelection!"); } @@ -188,11 +340,11 @@ void TargetLowering::AdjustInstrPostInstrSelection(MachineInstr *MI, // SelectionDAGISel code //===----------------------------------------------------------------------===// -SelectionDAGISel::SelectionDAGISel(const TargetMachine &tm, +SelectionDAGISel::SelectionDAGISel(TargetMachine &tm, CodeGenOpt::Level OL) : - MachineFunctionPass(ID), TM(tm), TLI(*tm.getTargetLowering()), - FuncInfo(new FunctionLoweringInfo(TLI)), - CurDAG(new SelectionDAG(tm)), + MachineFunctionPass(ID), TM(tm), + FuncInfo(new FunctionLoweringInfo()), + CurDAG(new SelectionDAG(tm, OL)), SDB(new SelectionDAGBuilder(*CurDAG, *FuncInfo, OL)), GFI(), OptLevel(OL), @@ -200,6 +352,8 @@ SelectionDAGISel::SelectionDAGISel(const TargetMachine &tm, initializeGCModuleInfoPass(*PassRegistry::getPassRegistry()); initializeAliasAnalysisAnalysisGroup(*PassRegistry::getPassRegistry()); initializeBranchProbabilityInfoPass(*PassRegistry::getPassRegistry()); + initializeTargetLibraryInfoWrapperPassPass( + *PassRegistry::getPassRegistry()); } SelectionDAGISel::~SelectionDAGISel() { @@ -213,6 +367,7 @@ void SelectionDAGISel::getAnalysisUsage(AnalysisUsage &AU) const { AU.addPreserved(); AU.addRequired(); AU.addPreserved(); + AU.addRequired(); if (UseMBPI && OptLevel != CodeGenOpt::None) AU.addRequired(); MachineFunctionPass::getAnalysisUsage(AU); @@ -225,11 +380,11 @@ void SelectionDAGISel::getAnalysisUsage(AnalysisUsage &AU) const { /// /// This is required for correctness, so it must be done at -O0. /// -static void SplitCriticalSideEffectEdges(Function &Fn, Pass *SDISel) { +static void SplitCriticalSideEffectEdges(Function &Fn, AliasAnalysis *AA) { // Loop for blocks with phi nodes. for (Function::iterator BB = Fn.begin(), E = Fn.end(); BB != E; ++BB) { PHINode *PN = dyn_cast(BB->begin()); - if (PN == 0) continue; + if (!PN) continue; ReprocessBlock: // For each block with a PHI node, check to see if any of the input values @@ -239,7 +394,7 @@ static void SplitCriticalSideEffectEdges(Function &Fn, Pass *SDISel) { for (BasicBlock::iterator I = BB->begin(); (PN = dyn_cast(I)); ++I) for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) { ConstantExpr *CE = dyn_cast(PN->getIncomingValue(i)); - if (CE == 0 || !CE->canTrap()) continue; + if (!CE || !CE->canTrap()) continue; // The only case we have to worry about is when the edge is critical. // Since this block has a PHI Node, we assume it has multiple input @@ -249,8 +404,9 @@ static void SplitCriticalSideEffectEdges(Function &Fn, Pass *SDISel) { continue; // Okay, we have to split this edge. - SplitCriticalEdge(Pred->getTerminator(), - GetSuccessorNumber(Pred, BB), SDISel, true); + SplitCriticalEdge( + Pred->getTerminator(), GetSuccessorNumber(Pred, BB), + CriticalEdgeSplittingOptions(AA).setMergeIdenticalEdges()); goto ReprocessBlock; } } @@ -258,33 +414,48 @@ static void SplitCriticalSideEffectEdges(Function &Fn, Pass *SDISel) { bool SelectionDAGISel::runOnMachineFunction(MachineFunction &mf) { // Do some sanity-checking on the command-line options. - assert((!EnableFastISelVerbose || EnableFastISel) && + assert((!EnableFastISelVerbose || TM.Options.EnableFastISel) && "-fast-isel-verbose requires -fast-isel"); - assert((!EnableFastISelAbort || EnableFastISel) && - "-fast-isel-abort requires -fast-isel"); + assert((!EnableFastISelAbort || TM.Options.EnableFastISel) && + "-fast-isel-abort > 0 requires -fast-isel"); const Function &Fn = *mf.getFunction(); - const TargetInstrInfo &TII = *TM.getInstrInfo(); - const TargetRegisterInfo &TRI = *TM.getRegisterInfo(); - MF = &mf; + + // Reset the target options before resetting the optimization + // level below. + // FIXME: This is a horrible hack and should be processed via + // codegen looking at the optimization level explicitly when + // it wants to look at it. + TM.resetTargetOptions(Fn); + // Reset OptLevel to None for optnone functions. + CodeGenOpt::Level NewOptLevel = OptLevel; + if (Fn.hasFnAttribute(Attribute::OptimizeNone)) + NewOptLevel = CodeGenOpt::None; + OptLevelChanger OLC(*this, NewOptLevel); + + TII = MF->getSubtarget().getInstrInfo(); + TLI = MF->getSubtarget().getTargetLowering(); RegInfo = &MF->getRegInfo(); AA = &getAnalysis(); - GFI = Fn.hasGC() ? &getAnalysis().getFunctionInfo(Fn) : 0; + LibInfo = &getAnalysis().getTLI(); + GFI = Fn.hasGC() ? &getAnalysis().getFunctionInfo(Fn) : nullptr; DEBUG(dbgs() << "\n\n\n=== " << Fn.getName() << "\n"); - SplitCriticalSideEffectEdges(const_cast(Fn), this); + SplitCriticalSideEffectEdges(const_cast(Fn), AA); CurDAG->init(*MF); - FuncInfo->set(Fn, *MF); + FuncInfo->set(Fn, *MF, CurDAG); if (UseMBPI && OptLevel != CodeGenOpt::None) FuncInfo->BPI = &getAnalysis(); else - FuncInfo->BPI = 0; + FuncInfo->BPI = nullptr; + + SDB->init(GFI, *AA, LibInfo); - SDB->init(GFI, *AA); + MF->setHasInlineAsm(false); SelectAllBasicBlocks(Fn); @@ -292,7 +463,8 @@ bool SelectionDAGISel::runOnMachineFunction(MachineFunction &mf) { // copied into vregs, emit the copies into the top of the block before // emitting the code for the block. MachineBasicBlock *EntryMBB = MF->begin(); - RegInfo->EmitLiveInCopies(EntryMBB, TRI, TII); + const TargetRegisterInfo &TRI = *MF->getSubtarget().getRegisterInfo(); + RegInfo->EmitLiveInCopies(EntryMBB, TRI, *TII); DenseMap LiveInMap; if (!FuncInfo->ArgDbgValues.empty()) @@ -304,77 +476,87 @@ bool SelectionDAGISel::runOnMachineFunction(MachineFunction &mf) { // Insert DBG_VALUE instructions for function arguments to the entry block. for (unsigned i = 0, e = FuncInfo->ArgDbgValues.size(); i != e; ++i) { MachineInstr *MI = FuncInfo->ArgDbgValues[e-i-1]; - unsigned Reg = MI->getOperand(0).getReg(); + bool hasFI = MI->getOperand(0).isFI(); + unsigned Reg = + hasFI ? TRI.getFrameRegister(*MF) : MI->getOperand(0).getReg(); if (TargetRegisterInfo::isPhysicalRegister(Reg)) EntryMBB->insert(EntryMBB->begin(), MI); else { MachineInstr *Def = RegInfo->getVRegDef(Reg); - MachineBasicBlock::iterator InsertPos = Def; - // FIXME: VR def may not be in entry block. - Def->getParent()->insert(llvm::next(InsertPos), MI); + if (Def) { + MachineBasicBlock::iterator InsertPos = Def; + // FIXME: VR def may not be in entry block. + Def->getParent()->insert(std::next(InsertPos), MI); + } else + DEBUG(dbgs() << "Dropping debug info for dead vreg" + << TargetRegisterInfo::virtReg2Index(Reg) << "\n"); } // If Reg is live-in then update debug info to track its copy in a vreg. DenseMap::iterator LDI = LiveInMap.find(Reg); if (LDI != LiveInMap.end()) { + assert(!hasFI && "There's no handling of frame pointer updating here yet " + "- add if needed"); MachineInstr *Def = RegInfo->getVRegDef(LDI->second); MachineBasicBlock::iterator InsertPos = Def; - const MDNode *Variable = - MI->getOperand(MI->getNumOperands()-1).getMetadata(); - unsigned Offset = MI->getOperand(1).getImm(); + const MDNode *Variable = MI->getDebugVariable(); + const MDNode *Expr = MI->getDebugExpression(); + DebugLoc DL = MI->getDebugLoc(); + bool IsIndirect = MI->isIndirectDebugValue(); + unsigned Offset = IsIndirect ? MI->getOperand(1).getImm() : 0; + assert(cast(Variable)->isValidLocationForIntrinsic(DL) && + "Expected inlined-at fields to agree"); // Def is never a terminator here, so it is ok to increment InsertPos. - BuildMI(*EntryMBB, ++InsertPos, MI->getDebugLoc(), - TII.get(TargetOpcode::DBG_VALUE)) - .addReg(LDI->second, RegState::Debug) - .addImm(Offset).addMetadata(Variable); + BuildMI(*EntryMBB, ++InsertPos, DL, TII->get(TargetOpcode::DBG_VALUE), + IsIndirect, LDI->second, Offset, Variable, Expr); // If this vreg is directly copied into an exported register then // that COPY instructions also need DBG_VALUE, if it is the only // user of LDI->second. - MachineInstr *CopyUseMI = NULL; - for (MachineRegisterInfo::use_iterator - UI = RegInfo->use_begin(LDI->second); - MachineInstr *UseMI = UI.skipInstruction();) { + MachineInstr *CopyUseMI = nullptr; + for (MachineRegisterInfo::use_instr_iterator + UI = RegInfo->use_instr_begin(LDI->second), + E = RegInfo->use_instr_end(); UI != E; ) { + MachineInstr *UseMI = &*(UI++); if (UseMI->isDebugValue()) continue; if (UseMI->isCopy() && !CopyUseMI && UseMI->getParent() == EntryMBB) { CopyUseMI = UseMI; continue; } // Otherwise this is another use or second copy use. - CopyUseMI = NULL; break; + CopyUseMI = nullptr; break; } if (CopyUseMI) { + // Use MI's debug location, which describes where Variable was + // declared, rather than whatever is attached to CopyUseMI. MachineInstr *NewMI = - BuildMI(*MF, CopyUseMI->getDebugLoc(), - TII.get(TargetOpcode::DBG_VALUE)) - .addReg(CopyUseMI->getOperand(0).getReg(), RegState::Debug) - .addImm(Offset).addMetadata(Variable); - EntryMBB->insertAfter(CopyUseMI, NewMI); + BuildMI(*MF, DL, TII->get(TargetOpcode::DBG_VALUE), IsIndirect, + CopyUseMI->getOperand(0).getReg(), Offset, Variable, Expr); + MachineBasicBlock::iterator Pos = CopyUseMI; + EntryMBB->insertAfter(Pos, NewMI); } } } // Determine if there are any calls in this machine function. MachineFrameInfo *MFI = MF->getFrameInfo(); - if (!MFI->hasCalls()) { - for (MachineFunction::const_iterator - I = MF->begin(), E = MF->end(); I != E; ++I) { - const MachineBasicBlock *MBB = I; - for (MachineBasicBlock::const_iterator - II = MBB->begin(), IE = MBB->end(); II != IE; ++II) { - const MCInstrDesc &MCID = TM.getInstrInfo()->get(II->getOpcode()); - - if ((MCID.isCall() && !MCID.isReturn()) || - II->isStackAligningInlineAsm()) { - MFI->setHasCalls(true); - goto done; - } + for (const auto &MBB : *MF) { + if (MFI->hasCalls() && MF->hasInlineAsm()) + break; + + for (const auto &MI : MBB) { + const MCInstrDesc &MCID = TII->get(MI.getOpcode()); + if ((MCID.isCall() && !MCID.isReturn()) || + MI.isStackAligningInlineAsm()) { + MFI->setHasCalls(true); + } + if (MI.isInlineAsm()) { + MF->setHasInlineAsm(true); } } - done:; } // Determine if there is a call to setjmp in the machine function. - MF->setCallsSetJmp(Fn.callsFunctionThatReturnsTwice()); + MF->setExposesReturnsTwice(Fn.callsFunctionThatReturnsTwice()); // Replace forward-declared registers with the registers containing // the desired value. @@ -387,28 +569,38 @@ bool SelectionDAGISel::runOnMachineFunction(MachineFunction &mf) { // If To is also scheduled to be replaced, find what its ultimate // replacement is. for (;;) { - DenseMap::iterator J = - FuncInfo->RegFixups.find(To); + DenseMap::iterator J = FuncInfo->RegFixups.find(To); if (J == E) break; To = J->second; } + // Make sure the new register has a sufficiently constrained register class. + if (TargetRegisterInfo::isVirtualRegister(From) && + TargetRegisterInfo::isVirtualRegister(To)) + MRI.constrainRegClass(To, MRI.getRegClass(From)); // Replace it. MRI.replaceRegWith(From, To); } + // Freeze the set of reserved registers now that MachineFrameInfo has been + // set up. All the information required by getReservedRegs() should be + // available now. + MRI.freezeReservedRegs(*MF); + // Release function-specific state. SDB and CurDAG are already cleared // at this point. FuncInfo->clear(); + DEBUG(dbgs() << "*** MachineFunction at end of ISel ***\n"); + DEBUG(MF->print(dbgs())); + return true; } void SelectionDAGISel::SelectBasicBlock(BasicBlock::const_iterator Begin, BasicBlock::const_iterator End, bool &HadTailCall) { - // Lower all of the non-terminator instructions. If a call is emitted - // as a tail call, cease emitting nodes for this block. Terminators - // are handled below. + // Lower the instructions. If a call is emitted as a tail call, cease emitting + // nodes for this block. for (BasicBlock::const_iterator I = Begin; I != End && !SDB->HasTailCall; ++I) SDB->visit(*I); @@ -427,7 +619,6 @@ void SelectionDAGISel::ComputeLiveOutVRegInfo() { Worklist.push_back(CurDAG->getRoot().getNode()); - APInt Mask; APInt KnownZero; APInt KnownOne; @@ -435,7 +626,7 @@ void SelectionDAGISel::ComputeLiveOutVRegInfo() { SDNode *N = Worklist.pop_back_val(); // If we've already seen this node, ignore it. - if (!VisitedNodes.insert(N)) + if (!VisitedNodes.insert(N).second) continue; // Otherwise, add all chain operands to the worklist. @@ -458,8 +649,7 @@ void SelectionDAGISel::ComputeLiveOutVRegInfo() { continue; unsigned NumSignBits = CurDAG->ComputeNumSignBits(Src); - Mask = APInt::getAllOnesValue(SrcVT.getSizeInBits()); - CurDAG->ComputeMaskedBits(Src, Mask, KnownZero, KnownOne); + CurDAG->computeKnownBits(Src, KnownZero, KnownOne); FuncInfo->AddLiveOutRegInfo(DestReg, NumSignBits, KnownZero, KnownOne); } while (!Worklist.empty()); } @@ -471,6 +661,12 @@ void SelectionDAGISel::CodeGenAndEmitDAG() { std::string BlockName; int BlockNumber = -1; (void)BlockNumber; + bool MatchFilterBB = false; (void)MatchFilterBB; +#ifndef NDEBUG + MatchFilterBB = (FilterDAGBasicBlockName.empty() || + FilterDAGBasicBlockName == + FuncInfo->MBB->getBasicBlock()->getName().str()); +#endif #ifdef NDEBUG if (ViewDAGCombine1 || ViewLegalizeTypesDAGs || ViewLegalizeDAGs || ViewDAGCombine2 || ViewDAGCombineLT || ViewISelDAGs || ViewSchedDAGs || @@ -478,18 +674,19 @@ void SelectionDAGISel::CodeGenAndEmitDAG() { #endif { BlockNumber = FuncInfo->MBB->getNumber(); - BlockName = MF->getFunction()->getNameStr() + ":" + - FuncInfo->MBB->getBasicBlock()->getNameStr(); + BlockName = + (MF->getName() + ":" + FuncInfo->MBB->getBasicBlock()->getName()).str(); } DEBUG(dbgs() << "Initial selection DAG: BB#" << BlockNumber << " '" << BlockName << "'\n"; CurDAG->dump()); - if (ViewDAGCombine1) CurDAG->viewGraph("dag-combine1 input for " + BlockName); + if (ViewDAGCombine1 && MatchFilterBB) + CurDAG->viewGraph("dag-combine1 input for " + BlockName); // Run the DAG combiner in pre-legalize mode. { NamedRegionTimer T("DAG Combining 1", GroupName, TimePassesIsEnabled); - CurDAG->Combine(Unrestricted, *AA, OptLevel); + CurDAG->Combine(BeforeLegalizeTypes, *AA, OptLevel); } DEBUG(dbgs() << "Optimized lowered selection DAG: BB#" << BlockNumber @@ -497,8 +694,8 @@ void SelectionDAGISel::CodeGenAndEmitDAG() { // Second step, hack on the DAG until it only uses operations and types that // the target supports. - if (ViewLegalizeTypesDAGs) CurDAG->viewGraph("legalize-types input for " + - BlockName); + if (ViewLegalizeTypesDAGs && MatchFilterBB) + CurDAG->viewGraph("legalize-types input for " + BlockName); bool Changed; { @@ -509,19 +706,22 @@ void SelectionDAGISel::CodeGenAndEmitDAG() { DEBUG(dbgs() << "Type-legalized selection DAG: BB#" << BlockNumber << " '" << BlockName << "'\n"; CurDAG->dump()); + CurDAG->NewNodesMustHaveLegalTypes = true; + if (Changed) { - if (ViewDAGCombineLT) + if (ViewDAGCombineLT && MatchFilterBB) CurDAG->viewGraph("dag-combine-lt input for " + BlockName); // Run the DAG combiner in post-type-legalize mode. { NamedRegionTimer T("DAG Combining after legalize types", GroupName, TimePassesIsEnabled); - CurDAG->Combine(NoIllegalTypes, *AA, OptLevel); + CurDAG->Combine(AfterLegalizeTypes, *AA, OptLevel); } DEBUG(dbgs() << "Optimized type-legalized selection DAG: BB#" << BlockNumber << " '" << BlockName << "'\n"; CurDAG->dump()); + } { @@ -535,21 +735,22 @@ void SelectionDAGISel::CodeGenAndEmitDAG() { CurDAG->LegalizeTypes(); } - if (ViewDAGCombineLT) + if (ViewDAGCombineLT && MatchFilterBB) CurDAG->viewGraph("dag-combine-lv input for " + BlockName); // Run the DAG combiner in post-type-legalize mode. { NamedRegionTimer T("DAG Combining after legalize vectors", GroupName, TimePassesIsEnabled); - CurDAG->Combine(NoIllegalOperations, *AA, OptLevel); + CurDAG->Combine(AfterLegalizeVectorOps, *AA, OptLevel); } DEBUG(dbgs() << "Optimized vector-legalized selection DAG: BB#" << BlockNumber << " '" << BlockName << "'\n"; CurDAG->dump()); } - if (ViewLegalizeDAGs) CurDAG->viewGraph("legalize input for " + BlockName); + if (ViewLegalizeDAGs && MatchFilterBB) + CurDAG->viewGraph("legalize input for " + BlockName); { NamedRegionTimer T("DAG Legalization", GroupName, TimePassesIsEnabled); @@ -559,12 +760,13 @@ void SelectionDAGISel::CodeGenAndEmitDAG() { DEBUG(dbgs() << "Legalized selection DAG: BB#" << BlockNumber << " '" << BlockName << "'\n"; CurDAG->dump()); - if (ViewDAGCombine2) CurDAG->viewGraph("dag-combine2 input for " + BlockName); + if (ViewDAGCombine2 && MatchFilterBB) + CurDAG->viewGraph("dag-combine2 input for " + BlockName); // Run the DAG combiner in post-legalize mode. { NamedRegionTimer T("DAG Combining 2", GroupName, TimePassesIsEnabled); - CurDAG->Combine(NoIllegalOperations, *AA, OptLevel); + CurDAG->Combine(AfterLegalizeDAG, *AA, OptLevel); } DEBUG(dbgs() << "Optimized legalized selection DAG: BB#" << BlockNumber @@ -573,7 +775,8 @@ void SelectionDAGISel::CodeGenAndEmitDAG() { if (OptLevel != CodeGenOpt::None) ComputeLiveOutVRegInfo(); - if (ViewISelDAGs) CurDAG->viewGraph("isel input for " + BlockName); + if (ViewISelDAGs && MatchFilterBB) + CurDAG->viewGraph("isel input for " + BlockName); // Third, instruction select all of the operations to machine code, adding the // code to the MachineBasicBlock. @@ -585,17 +788,18 @@ void SelectionDAGISel::CodeGenAndEmitDAG() { DEBUG(dbgs() << "Selected selection DAG: BB#" << BlockNumber << " '" << BlockName << "'\n"; CurDAG->dump()); - if (ViewSchedDAGs) CurDAG->viewGraph("scheduler input for " + BlockName); + if (ViewSchedDAGs && MatchFilterBB) + CurDAG->viewGraph("scheduler input for " + BlockName); // Schedule machine code. ScheduleDAGSDNodes *Scheduler = CreateScheduler(); { NamedRegionTimer T("Instruction Scheduling", GroupName, TimePassesIsEnabled); - Scheduler->Run(CurDAG, FuncInfo->MBB, FuncInfo->InsertPt); + Scheduler->Run(CurDAG, FuncInfo->MBB); } - if (ViewSUnitDAGs) Scheduler->viewGraph(); + if (ViewSUnitDAGs && MatchFilterBB) Scheduler->viewGraph(); // Emit machine code to BB. This can change 'BB' to the last block being // inserted into. @@ -603,8 +807,9 @@ void SelectionDAGISel::CodeGenAndEmitDAG() { { NamedRegionTimer T("Instruction Creation", GroupName, TimePassesIsEnabled); - LastMBB = FuncInfo->MBB = Scheduler->EmitSchedule(); - FuncInfo->InsertPt = Scheduler->InsertPos; + // FuncInfo->InsertPt is passed by reference and set to the end of the + // scheduled instructions. + LastMBB = FuncInfo->MBB = Scheduler->EmitSchedule(FuncInfo->InsertPt); } // If the block was split, make sure we update any references that are used to @@ -623,8 +828,27 @@ void SelectionDAGISel::CodeGenAndEmitDAG() { CurDAG->clear(); } +namespace { +/// ISelUpdater - helper class to handle updates of the instruction selection +/// graph. +class ISelUpdater : public SelectionDAG::DAGUpdateListener { + SelectionDAG::allnodes_iterator &ISelPosition; +public: + ISelUpdater(SelectionDAG &DAG, SelectionDAG::allnodes_iterator &isp) + : SelectionDAG::DAGUpdateListener(DAG), ISelPosition(isp) {} + + /// NodeDeleted - Handle nodes deleted from the graph. If the node being + /// deleted is the current ISelPosition node, update ISelPosition. + /// + void NodeDeleted(SDNode *N, SDNode *E) override { + if (ISelPosition == SelectionDAG::allnodes_iterator(N)) + ++ISelPosition; + } +}; +} // end anonymous namespace + void SelectionDAGISel::DoInstructionSelection() { - DEBUG(errs() << "===== Instruction selection begins: BB#" + DEBUG(dbgs() << "===== Instruction selection begins: BB#" << FuncInfo->MBB->getNumber() << " '" << FuncInfo->MBB->getName() << "'\n"); @@ -639,9 +863,13 @@ void SelectionDAGISel::DoInstructionSelection() { // a reference to the root node, preventing it from being deleted, // and tracking any changes of the root. HandleSDNode Dummy(CurDAG->getRoot()); - ISelPosition = SelectionDAG::allnodes_iterator(CurDAG->getRoot().getNode()); + SelectionDAG::allnodes_iterator ISelPosition (CurDAG->getRoot().getNode()); ++ISelPosition; + // Make sure that ISelPosition gets properly updated when nodes are deleted + // in calls made from this function. + ISelUpdater ISU(*CurDAG, ISelPosition); + // The AllNodes list is now topological-sorted. Visit the // nodes by starting at the end of the list (the root of the // graph) and preceding back toward the beginning (the entry @@ -663,151 +891,89 @@ void SelectionDAGISel::DoInstructionSelection() { if (ResNode == Node || Node->getOpcode() == ISD::DELETED_NODE) continue; // Replace node. - if (ResNode) + if (ResNode) { ReplaceUses(Node, ResNode); + } // If after the replacement this node is not used any more, // remove this dead node. - if (Node->use_empty()) { // Don't delete EntryToken, etc. - ISelUpdater ISU(ISelPosition); - CurDAG->RemoveDeadNode(Node, &ISU); - } + if (Node->use_empty()) // Don't delete EntryToken, etc. + CurDAG->RemoveDeadNode(Node); } CurDAG->setRoot(Dummy.getValue()); } - DEBUG(errs() << "===== Instruction selection ends:\n"); + DEBUG(dbgs() << "===== Instruction selection ends:\n"); PostprocessISelDAG(); } /// PrepareEHLandingPad - Emit an EH_LABEL, set up live-in registers, and /// do other setup for EH landing-pad blocks. -void SelectionDAGISel::PrepareEHLandingPad() { +bool SelectionDAGISel::PrepareEHLandingPad() { MachineBasicBlock *MBB = FuncInfo->MBB; + const TargetRegisterClass *PtrRC = TLI->getRegClassFor(TLI->getPointerTy()); + // Add a label to mark the beginning of the landing pad. Deletion of the // landing pad can thus be detected via the MachineModuleInfo. MCSymbol *Label = MF->getMMI().addLandingPad(MBB); // Assign the call site to the landing pad's begin label. MF->getMMI().setCallSiteLandingPad(Label, SDB->LPadToCallSiteMap[MBB]); - - const MCInstrDesc &II = TM.getInstrInfo()->get(TargetOpcode::EH_LABEL); + + const MCInstrDesc &II = TII->get(TargetOpcode::EH_LABEL); BuildMI(*MBB, FuncInfo->InsertPt, SDB->getCurDebugLoc(), II) .addSym(Label); - // Mark exception register as live in. - unsigned Reg = TLI.getExceptionAddressRegister(); - if (Reg) MBB->addLiveIn(Reg); - - // Mark exception selector register as live in. - Reg = TLI.getExceptionSelectorRegister(); - if (Reg) MBB->addLiveIn(Reg); - - // FIXME: Hack around an exception handling flaw (PR1508): the personality - // function and list of typeids logically belong to the invoke (or, if you - // like, the basic block containing the invoke), and need to be associated - // with it in the dwarf exception handling tables. Currently however the - // information is provided by an intrinsic (eh.selector) that can be moved - // to unexpected places by the optimizers: if the unwind edge is critical, - // then breaking it can result in the intrinsics being in the successor of - // the landing pad, not the landing pad itself. This results - // in exceptions not being caught because no typeids are associated with - // the invoke. This may not be the only way things can go wrong, but it - // is the only way we try to work around for the moment. + // If this is an MSVC-style personality function, we need to split the landing + // pad into several BBs. const BasicBlock *LLVMBB = MBB->getBasicBlock(); - const BranchInst *Br = dyn_cast(LLVMBB->getTerminator()); - - if (Br && Br->isUnconditional()) { // Critical edge? - BasicBlock::const_iterator I, E; - for (I = LLVMBB->begin(), E = --LLVMBB->end(); I != E; ++I) - if (isa(I)) - break; - - if (I == E) - // No catch info found - try to extract some from the successor. - CopyCatchInfo(Br->getSuccessor(0), LLVMBB, &MF->getMMI(), *FuncInfo); - } -} - -/// TryToFoldFastISelLoad - We're checking to see if we can fold the specified -/// load into the specified FoldInst. Note that we could have a sequence where -/// multiple LLVM IR instructions are folded into the same machineinstr. For -/// example we could have: -/// A: x = load i32 *P -/// B: y = icmp A, 42 -/// C: br y, ... -/// -/// In this scenario, LI is "A", and FoldInst is "C". We know about "B" (and -/// any other folded instructions) because it is between A and C. -/// -/// If we succeed in folding the load into the operation, return true. -/// -bool SelectionDAGISel::TryToFoldFastISelLoad(const LoadInst *LI, - const Instruction *FoldInst, - FastISel *FastIS) { - // We know that the load has a single use, but don't know what it is. If it - // isn't one of the folded instructions, then we can't succeed here. Handle - // this by scanning the single-use users of the load until we get to FoldInst. - unsigned MaxUsers = 6; // Don't scan down huge single-use chains of instrs. - - const Instruction *TheUser = LI->use_back(); - while (TheUser != FoldInst && // Scan up until we find FoldInst. - // Stay in the right block. - TheUser->getParent() == FoldInst->getParent() && - --MaxUsers) { // Don't scan too far. - // If there are multiple or no uses of this instruction, then bail out. - if (!TheUser->hasOneUse()) - return false; - - TheUser = TheUser->use_back(); - } - - // If we didn't find the fold instruction, then we failed to collapse the - // sequence. - if (TheUser != FoldInst) - return false; - - // Don't try to fold volatile loads. Target has to deal with alignment - // constraints. - if (LI->isVolatile()) return false; - - // Figure out which vreg this is going into. If there is no assigned vreg yet - // then there actually was no reference to it. Perhaps the load is referenced - // by a dead instruction. - unsigned LoadReg = FastIS->getRegForValue(LI); - if (LoadReg == 0) - return false; + const LandingPadInst *LPadInst = LLVMBB->getLandingPadInst(); + MF->getMMI().addPersonality( + MBB, cast(LPadInst->getPersonalityFn()->stripPointerCasts())); + EHPersonality Personality = MF->getMMI().getPersonalityType(); + + if (isMSVCEHPersonality(Personality)) { + SmallVector ClauseBBs; + const IntrinsicInst *ActionsCall = + dyn_cast(LLVMBB->getFirstInsertionPt()); + // Get all invoke BBs that unwind to this landingpad. + SmallVector InvokeBBs(MBB->pred_begin(), + MBB->pred_end()); + if (ActionsCall && ActionsCall->getIntrinsicID() == Intrinsic::eh_actions) { + // If this is a call to llvm.eh.actions followed by indirectbr, then we've + // run WinEHPrepare, and we should remove this block from the machine CFG. + // Mark the targets of the indirectbr as landingpads instead. + for (const BasicBlock *LLVMSucc : successors(LLVMBB)) { + MachineBasicBlock *ClauseBB = FuncInfo->MBBMap[LLVMSucc]; + // Add the edge from the invoke to the clause. + for (MachineBasicBlock *InvokeBB : InvokeBBs) + InvokeBB->addSuccessor(ClauseBB); + + // Mark the clause as a landing pad or MI passes will delete it. + ClauseBB->setIsLandingPad(); + } + } - // Check to see what the uses of this vreg are. If it has no uses, or more - // than one use (at the machine instr level) then we can't fold it. - MachineRegisterInfo::reg_iterator RI = RegInfo->reg_begin(LoadReg); - if (RI == RegInfo->reg_end()) - return false; + // Remove the edge from the invoke to the lpad. + for (MachineBasicBlock *InvokeBB : InvokeBBs) + InvokeBB->removeSuccessor(MBB); - // See if there is exactly one use of the vreg. If there are multiple uses, - // then the instruction got lowered to multiple machine instructions or the - // use of the loaded value ended up being multiple operands of the result, in - // either case, we can't fold this. - MachineRegisterInfo::reg_iterator PostRI = RI; ++PostRI; - if (PostRI != RegInfo->reg_end()) + // Don't select instructions for the landingpad. return false; + } - assert(RI.getOperand().isUse() && - "The only use of the vreg must be a use, we haven't emitted the def!"); - - MachineInstr *User = &*RI; + // Mark exception register as live in. + if (unsigned Reg = TLI->getExceptionPointerRegister()) + FuncInfo->ExceptionPointerVirtReg = MBB->addLiveIn(Reg, PtrRC); - // Set the insertion point properly. Folding the load can cause generation of - // other random instructions (like sign extends) for addressing modes, make - // sure they get inserted in a logical place before the new instruction. - FuncInfo->InsertPt = User; - FuncInfo->MBB = User->getParent(); + // Mark exception selector register as live in. + if (unsigned Reg = TLI->getExceptionSelectorRegister()) + FuncInfo->ExceptionSelectorVirtReg = MBB->addLiveIn(Reg, PtrRC); - // Ask the target to try folding the load. - return FastIS->TryToFoldLoad(User, RI.getOperandNo(), LI); + return true; } /// isFoldedOrDeadInstruction - Return true if the specified instruction is @@ -822,11 +988,121 @@ static bool isFoldedOrDeadInstruction(const Instruction *I, !FuncInfo->isExportedInst(I); // Exported instrs must be computed. } +#ifndef NDEBUG +// Collect per Instruction statistics for fast-isel misses. Only those +// instructions that cause the bail are accounted for. It does not account for +// instructions higher in the block. Thus, summing the per instructions stats +// will not add up to what is reported by NumFastIselFailures. +static void collectFailStats(const Instruction *I) { + switch (I->getOpcode()) { + default: assert (0 && " "); + + // Terminators + case Instruction::Ret: NumFastIselFailRet++; return; + case Instruction::Br: NumFastIselFailBr++; return; + case Instruction::Switch: NumFastIselFailSwitch++; return; + case Instruction::IndirectBr: NumFastIselFailIndirectBr++; return; + case Instruction::Invoke: NumFastIselFailInvoke++; return; + case Instruction::Resume: NumFastIselFailResume++; return; + case Instruction::Unreachable: NumFastIselFailUnreachable++; return; + + // Standard binary operators... + case Instruction::Add: NumFastIselFailAdd++; return; + case Instruction::FAdd: NumFastIselFailFAdd++; return; + case Instruction::Sub: NumFastIselFailSub++; return; + case Instruction::FSub: NumFastIselFailFSub++; return; + case Instruction::Mul: NumFastIselFailMul++; return; + case Instruction::FMul: NumFastIselFailFMul++; return; + case Instruction::UDiv: NumFastIselFailUDiv++; return; + case Instruction::SDiv: NumFastIselFailSDiv++; return; + case Instruction::FDiv: NumFastIselFailFDiv++; return; + case Instruction::URem: NumFastIselFailURem++; return; + case Instruction::SRem: NumFastIselFailSRem++; return; + case Instruction::FRem: NumFastIselFailFRem++; return; + + // Logical operators... + case Instruction::And: NumFastIselFailAnd++; return; + case Instruction::Or: NumFastIselFailOr++; return; + case Instruction::Xor: NumFastIselFailXor++; return; + + // Memory instructions... + case Instruction::Alloca: NumFastIselFailAlloca++; return; + case Instruction::Load: NumFastIselFailLoad++; return; + case Instruction::Store: NumFastIselFailStore++; return; + case Instruction::AtomicCmpXchg: NumFastIselFailAtomicCmpXchg++; return; + case Instruction::AtomicRMW: NumFastIselFailAtomicRMW++; return; + case Instruction::Fence: NumFastIselFailFence++; return; + case Instruction::GetElementPtr: NumFastIselFailGetElementPtr++; return; + + // Convert instructions... + case Instruction::Trunc: NumFastIselFailTrunc++; return; + case Instruction::ZExt: NumFastIselFailZExt++; return; + case Instruction::SExt: NumFastIselFailSExt++; return; + case Instruction::FPTrunc: NumFastIselFailFPTrunc++; return; + case Instruction::FPExt: NumFastIselFailFPExt++; return; + case Instruction::FPToUI: NumFastIselFailFPToUI++; return; + case Instruction::FPToSI: NumFastIselFailFPToSI++; return; + case Instruction::UIToFP: NumFastIselFailUIToFP++; return; + case Instruction::SIToFP: NumFastIselFailSIToFP++; return; + case Instruction::IntToPtr: NumFastIselFailIntToPtr++; return; + case Instruction::PtrToInt: NumFastIselFailPtrToInt++; return; + case Instruction::BitCast: NumFastIselFailBitCast++; return; + + // Other instructions... + case Instruction::ICmp: NumFastIselFailICmp++; return; + case Instruction::FCmp: NumFastIselFailFCmp++; return; + case Instruction::PHI: NumFastIselFailPHI++; return; + case Instruction::Select: NumFastIselFailSelect++; return; + case Instruction::Call: { + if (auto const *Intrinsic = dyn_cast(I)) { + switch (Intrinsic->getIntrinsicID()) { + default: + NumFastIselFailIntrinsicCall++; return; + case Intrinsic::sadd_with_overflow: + NumFastIselFailSAddWithOverflow++; return; + case Intrinsic::uadd_with_overflow: + NumFastIselFailUAddWithOverflow++; return; + case Intrinsic::ssub_with_overflow: + NumFastIselFailSSubWithOverflow++; return; + case Intrinsic::usub_with_overflow: + NumFastIselFailUSubWithOverflow++; return; + case Intrinsic::smul_with_overflow: + NumFastIselFailSMulWithOverflow++; return; + case Intrinsic::umul_with_overflow: + NumFastIselFailUMulWithOverflow++; return; + case Intrinsic::frameaddress: + NumFastIselFailFrameaddress++; return; + case Intrinsic::sqrt: + NumFastIselFailSqrt++; return; + case Intrinsic::experimental_stackmap: + NumFastIselFailStackMap++; return; + case Intrinsic::experimental_patchpoint_void: // fall-through + case Intrinsic::experimental_patchpoint_i64: + NumFastIselFailPatchPoint++; return; + } + } + NumFastIselFailCall++; + return; + } + case Instruction::Shl: NumFastIselFailShl++; return; + case Instruction::LShr: NumFastIselFailLShr++; return; + case Instruction::AShr: NumFastIselFailAShr++; return; + case Instruction::VAArg: NumFastIselFailVAArg++; return; + case Instruction::ExtractElement: NumFastIselFailExtractElement++; return; + case Instruction::InsertElement: NumFastIselFailInsertElement++; return; + case Instruction::ShuffleVector: NumFastIselFailShuffleVector++; return; + case Instruction::ExtractValue: NumFastIselFailExtractValue++; return; + case Instruction::InsertValue: NumFastIselFailInsertValue++; return; + case Instruction::LandingPad: NumFastIselFailLandingPad++; return; + } +} +#endif + void SelectionDAGISel::SelectAllBasicBlocks(const Function &Fn) { // Initialize the Fast-ISel state, if needed. - FastISel *FastIS = 0; - if (EnableFastISel) - FastIS = TLI.createFastISel(*FuncInfo); + FastISel *FastIS = nullptr; + if (TM.Options.EnableFastISel) + FastIS = TLI->createFastISel(*FuncInfo, LibInfo); // Iterate over all basic blocks in the function. ReversePostOrderTraversal RPOT(&Fn); @@ -846,33 +1122,30 @@ void SelectionDAGISel::SelectAllBasicBlocks(const Function &Fn) { if (AllPredsVisited) { for (BasicBlock::const_iterator I = LLVMBB->begin(); - isa(I); ++I) - FuncInfo->ComputePHILiveOutRegInfo(cast(I)); + const PHINode *PN = dyn_cast(I); ++I) + FuncInfo->ComputePHILiveOutRegInfo(PN); } else { for (BasicBlock::const_iterator I = LLVMBB->begin(); - isa(I); ++I) - FuncInfo->InvalidatePHILiveOutRegInfo(cast(I)); + const PHINode *PN = dyn_cast(I); ++I) + FuncInfo->InvalidatePHILiveOutRegInfo(PN); } FuncInfo->VisitedBBs.insert(LLVMBB); } - FuncInfo->MBB = FuncInfo->MBBMap[LLVMBB]; - FuncInfo->InsertPt = FuncInfo->MBB->getFirstNonPHI(); - BasicBlock::const_iterator const Begin = LLVMBB->getFirstNonPHI(); BasicBlock::const_iterator const End = LLVMBB->end(); BasicBlock::const_iterator BI = End; + FuncInfo->MBB = FuncInfo->MBBMap[LLVMBB]; FuncInfo->InsertPt = FuncInfo->MBB->getFirstNonPHI(); // Setup an EH landing-pad block. - if (FuncInfo->MBB->isLandingPad()) - PrepareEHLandingPad(); - - // Lower any arguments needed in this block if this is the entry block. - if (LLVMBB == &Fn.getEntryBlock()) - LowerArguments(LLVMBB); + FuncInfo->ExceptionPointerVirtReg = 0; + FuncInfo->ExceptionSelectorVirtReg = 0; + if (LLVMBB->isLandingPad()) + if (!PrepareEHLandingPad()) + continue; // Before doing SelectionDAG ISel, see if FastISel has been requested. if (FastIS) { @@ -881,58 +1154,86 @@ void SelectionDAGISel::SelectAllBasicBlocks(const Function &Fn) { // Emit code for any incoming arguments. This must happen before // beginning FastISel on the entry block. if (LLVMBB == &Fn.getEntryBlock()) { - CurDAG->setRoot(SDB->getControlRoot()); - SDB->clear(); - CodeGenAndEmitDAG(); + ++NumEntryBlocks; + + // Lower any arguments needed in this block if this is the entry block. + if (!FastIS->lowerArguments()) { + // Fast isel failed to lower these arguments + ++NumFastIselFailLowerArguments; + if (EnableFastISelAbort > 1) + report_fatal_error("FastISel didn't lower all arguments"); + + // Use SelectionDAG argument lowering + LowerArguments(Fn); + CurDAG->setRoot(SDB->getControlRoot()); + SDB->clear(); + CodeGenAndEmitDAG(); + } // If we inserted any instructions at the beginning, make a note of // where they are, so we can be sure to emit subsequent instructions // after them. if (FuncInfo->InsertPt != FuncInfo->MBB->begin()) - FastIS->setLastLocalValue(llvm::prior(FuncInfo->InsertPt)); + FastIS->setLastLocalValue(std::prev(FuncInfo->InsertPt)); else - FastIS->setLastLocalValue(0); + FastIS->setLastLocalValue(nullptr); } + unsigned NumFastIselRemaining = std::distance(Begin, End); // Do FastISel on as many instructions as possible. for (; BI != Begin; --BI) { - const Instruction *Inst = llvm::prior(BI); + const Instruction *Inst = std::prev(BI); // If we no longer require this instruction, skip it. - if (isFoldedOrDeadInstruction(Inst, FuncInfo)) + if (isFoldedOrDeadInstruction(Inst, FuncInfo)) { + --NumFastIselRemaining; continue; + } // Bottom-up: reset the insert pos at the top, after any local-value // instructions. FastIS->recomputeInsertPt(); // Try to select the instruction with FastISel. - if (FastIS->SelectInstruction(Inst)) { + if (FastIS->selectInstruction(Inst)) { + --NumFastIselRemaining; ++NumFastIselSuccess; // If fast isel succeeded, skip over all the folded instructions, and // then see if there is a load right before the selected instructions. // Try to fold the load if so. const Instruction *BeforeInst = Inst; while (BeforeInst != Begin) { - BeforeInst = llvm::prior(BasicBlock::const_iterator(BeforeInst)); + BeforeInst = std::prev(BasicBlock::const_iterator(BeforeInst)); if (!isFoldedOrDeadInstruction(BeforeInst, FuncInfo)) break; } if (BeforeInst != Inst && isa(BeforeInst) && BeforeInst->hasOneUse() && - TryToFoldFastISelLoad(cast(BeforeInst), Inst, FastIS)) + FastIS->tryToFoldLoad(cast(BeforeInst), Inst)) { // If we succeeded, don't re-select the load. - BI = llvm::next(BasicBlock::const_iterator(BeforeInst)); + BI = std::next(BasicBlock::const_iterator(BeforeInst)); + --NumFastIselRemaining; + ++NumFastIselSuccess; + } continue; } +#ifndef NDEBUG + if (EnableFastISelVerbose2) + collectFailStats(Inst); +#endif + // Then handle certain instructions as single-LLVM-Instruction blocks. if (isa(Inst)) { - ++NumFastIselFailures; + if (EnableFastISelVerbose || EnableFastISelAbort) { dbgs() << "FastISel missed call: "; Inst->dump(); } + if (EnableFastISelAbort > 2) + // FastISel selector couldn't handle something and bailed. + // For the purpose of debugging, just abort. + report_fatal_error("FastISel didn't select the entire block"); if (!Inst->getType()->isVoidTy() && !Inst->use_empty()) { unsigned &R = FuncInfo->ValueMap[Inst]; @@ -941,39 +1242,53 @@ void SelectionDAGISel::SelectAllBasicBlocks(const Function &Fn) { } bool HadTailCall = false; + MachineBasicBlock::iterator SavedInsertPt = FuncInfo->InsertPt; SelectBasicBlock(Inst, BI, HadTailCall); // If the call was emitted as a tail call, we're done with the block. + // We also need to delete any previously emitted instructions. if (HadTailCall) { + FastIS->removeDeadCode(SavedInsertPt, FuncInfo->MBB->end()); --BI; break; } + // Recompute NumFastIselRemaining as Selection DAG instruction + // selection may have handled the call, input args, etc. + unsigned RemainingNow = std::distance(Begin, BI); + NumFastIselFailures += NumFastIselRemaining - RemainingNow; + NumFastIselRemaining = RemainingNow; continue; } - if (isa(Inst) && !isa(Inst)) { - // Don't abort, and use a different message for terminator misses. - ++NumFastIselFailures; - if (EnableFastISelVerbose || EnableFastISelAbort) { + bool ShouldAbort = EnableFastISelAbort; + if (EnableFastISelVerbose || EnableFastISelAbort) { + if (isa(Inst)) { + // Use a different message for terminator misses. dbgs() << "FastISel missed terminator: "; - Inst->dump(); - } - } else { - ++NumFastIselFailures; - if (EnableFastISelVerbose || EnableFastISelAbort) { + // Don't abort unless for terminator unless the level is really high + ShouldAbort = (EnableFastISelAbort > 2); + } else { dbgs() << "FastISel miss: "; - Inst->dump(); } - if (EnableFastISelAbort) - // The "fast" selector couldn't handle something and bailed. - // For the purpose of debugging, just abort. - llvm_unreachable("FastISel didn't select the entire block"); + Inst->dump(); } + if (ShouldAbort) + // FastISel selector couldn't handle something and bailed. + // For the purpose of debugging, just abort. + report_fatal_error("FastISel didn't select the entire block"); + + NumFastIselFailures += NumFastIselRemaining; break; } FastIS->recomputeInsertPt(); + } else { + // Lower any arguments needed in this block if this is the entry block. + if (LLVMBB == &Fn.getEntryBlock()) { + ++NumEntryBlocks; + LowerArguments(Fn); + } } if (Begin != BI) @@ -995,6 +1310,91 @@ void SelectionDAGISel::SelectAllBasicBlocks(const Function &Fn) { delete FastIS; SDB->clearDanglingDebugInfo(); + SDB->SPDescriptor.resetPerFunctionState(); +} + +/// Given that the input MI is before a partial terminator sequence TSeq, return +/// true if M + TSeq also a partial terminator sequence. +/// +/// A Terminator sequence is a sequence of MachineInstrs which at this point in +/// lowering copy vregs into physical registers, which are then passed into +/// terminator instructors so we can satisfy ABI constraints. A partial +/// terminator sequence is an improper subset of a terminator sequence (i.e. it +/// may be the whole terminator sequence). +static bool MIIsInTerminatorSequence(const MachineInstr *MI) { + // If we do not have a copy or an implicit def, we return true if and only if + // MI is a debug value. + if (!MI->isCopy() && !MI->isImplicitDef()) + // Sometimes DBG_VALUE MI sneak in between the copies from the vregs to the + // physical registers if there is debug info associated with the terminator + // of our mbb. We want to include said debug info in our terminator + // sequence, so we return true in that case. + return MI->isDebugValue(); + + // We have left the terminator sequence if we are not doing one of the + // following: + // + // 1. Copying a vreg into a physical register. + // 2. Copying a vreg into a vreg. + // 3. Defining a register via an implicit def. + + // OPI should always be a register definition... + MachineInstr::const_mop_iterator OPI = MI->operands_begin(); + if (!OPI->isReg() || !OPI->isDef()) + return false; + + // Defining any register via an implicit def is always ok. + if (MI->isImplicitDef()) + return true; + + // Grab the copy source... + MachineInstr::const_mop_iterator OPI2 = OPI; + ++OPI2; + assert(OPI2 != MI->operands_end() + && "Should have a copy implying we should have 2 arguments."); + + // Make sure that the copy dest is not a vreg when the copy source is a + // physical register. + if (!OPI2->isReg() || + (!TargetRegisterInfo::isPhysicalRegister(OPI->getReg()) && + TargetRegisterInfo::isPhysicalRegister(OPI2->getReg()))) + return false; + + return true; +} + +/// Find the split point at which to splice the end of BB into its success stack +/// protector check machine basic block. +/// +/// On many platforms, due to ABI constraints, terminators, even before register +/// allocation, use physical registers. This creates an issue for us since +/// physical registers at this point can not travel across basic +/// blocks. Luckily, selectiondag always moves physical registers into vregs +/// when they enter functions and moves them through a sequence of copies back +/// into the physical registers right before the terminator creating a +/// ``Terminator Sequence''. This function is searching for the beginning of the +/// terminator sequence so that we can ensure that we splice off not just the +/// terminator, but additionally the copies that move the vregs into the +/// physical registers. +static MachineBasicBlock::iterator +FindSplitPointForStackProtector(MachineBasicBlock *BB, DebugLoc DL) { + MachineBasicBlock::iterator SplitPoint = BB->getFirstTerminator(); + // + if (SplitPoint == BB->begin()) + return SplitPoint; + + MachineBasicBlock::iterator Start = BB->begin(); + MachineBasicBlock::iterator Previous = SplitPoint; + --Previous; + + while (MIIsInTerminatorSequence(Previous)) { + SplitPoint = Previous; + if (Previous == Start) + break; + --Previous; + } + + return SplitPoint; } void @@ -1009,20 +1409,55 @@ SelectionDAGISel::FinishBasicBlock() { // Next, now that we know what the last MBB the LLVM BB expanded is, update // PHI nodes in successors. - if (SDB->SwitchCases.empty() && - SDB->JTCases.empty() && - SDB->BitTestCases.empty()) { - for (unsigned i = 0, e = FuncInfo->PHINodesToUpdate.size(); i != e; ++i) { - MachineInstr *PHI = FuncInfo->PHINodesToUpdate[i].first; - assert(PHI->isPHI() && - "This is not a machine PHI node that we are updating!"); - if (!FuncInfo->MBB->isSuccessor(PHI->getParent())) - continue; - PHI->addOperand( - MachineOperand::CreateReg(FuncInfo->PHINodesToUpdate[i].second, false)); - PHI->addOperand(MachineOperand::CreateMBB(FuncInfo->MBB)); + for (unsigned i = 0, e = FuncInfo->PHINodesToUpdate.size(); i != e; ++i) { + MachineInstrBuilder PHI(*MF, FuncInfo->PHINodesToUpdate[i].first); + assert(PHI->isPHI() && + "This is not a machine PHI node that we are updating!"); + if (!FuncInfo->MBB->isSuccessor(PHI->getParent())) + continue; + PHI.addReg(FuncInfo->PHINodesToUpdate[i].second).addMBB(FuncInfo->MBB); + } + + // Handle stack protector. + if (SDB->SPDescriptor.shouldEmitStackProtector()) { + MachineBasicBlock *ParentMBB = SDB->SPDescriptor.getParentMBB(); + MachineBasicBlock *SuccessMBB = SDB->SPDescriptor.getSuccessMBB(); + + // Find the split point to split the parent mbb. At the same time copy all + // physical registers used in the tail of parent mbb into virtual registers + // before the split point and back into physical registers after the split + // point. This prevents us needing to deal with Live-ins and many other + // register allocation issues caused by us splitting the parent mbb. The + // register allocator will clean up said virtual copies later on. + MachineBasicBlock::iterator SplitPoint = + FindSplitPointForStackProtector(ParentMBB, SDB->getCurDebugLoc()); + + // Splice the terminator of ParentMBB into SuccessMBB. + SuccessMBB->splice(SuccessMBB->end(), ParentMBB, + SplitPoint, + ParentMBB->end()); + + // Add compare/jump on neq/jump to the parent BB. + FuncInfo->MBB = ParentMBB; + FuncInfo->InsertPt = ParentMBB->end(); + SDB->visitSPDescriptorParent(SDB->SPDescriptor, ParentMBB); + CurDAG->setRoot(SDB->getRoot()); + SDB->clear(); + CodeGenAndEmitDAG(); + + // CodeGen Failure MBB if we have not codegened it yet. + MachineBasicBlock *FailureMBB = SDB->SPDescriptor.getFailureMBB(); + if (!FailureMBB->size()) { + FuncInfo->MBB = FailureMBB; + FuncInfo->InsertPt = FailureMBB->end(); + SDB->visitSPDescriptorFailure(SDB->SPDescriptor); + CurDAG->setRoot(SDB->getRoot()); + SDB->clear(); + CodeGenAndEmitDAG(); } - return; + + // Clear the Per-BB State. + SDB->SPDescriptor.resetPerBBState(); } for (unsigned i = 0, e = SDB->BitTestCases.size(); i != e; ++i) { @@ -1038,7 +1473,12 @@ SelectionDAGISel::FinishBasicBlock() { CodeGenAndEmitDAG(); } + uint32_t UnhandledWeight = 0; + for (unsigned j = 0, ej = SDB->BitTestCases[i].Cases.size(); j != ej; ++j) + UnhandledWeight += SDB->BitTestCases[i].Cases[j].ExtraWeight; + for (unsigned j = 0, ej = SDB->BitTestCases[i].Cases.size(); j != ej; ++j) { + UnhandledWeight -= SDB->BitTestCases[i].Cases[j].ExtraWeight; // Set the current basic block to the mbb we wish to insert the code into FuncInfo->MBB = SDB->BitTestCases[i].Cases[j].ThisBB; FuncInfo->InsertPt = FuncInfo->MBB->end(); @@ -1046,12 +1486,14 @@ SelectionDAGISel::FinishBasicBlock() { if (j+1 != ej) SDB->visitBitTestCase(SDB->BitTestCases[i], SDB->BitTestCases[i].Cases[j+1].ThisBB, + UnhandledWeight, SDB->BitTestCases[i].Reg, SDB->BitTestCases[i].Cases[j], FuncInfo->MBB); else SDB->visitBitTestCase(SDB->BitTestCases[i], SDB->BitTestCases[i].Default, + UnhandledWeight, SDB->BitTestCases[i].Reg, SDB->BitTestCases[i].Cases[j], FuncInfo->MBB); @@ -1065,33 +1507,23 @@ SelectionDAGISel::FinishBasicBlock() { // Update PHI Nodes for (unsigned pi = 0, pe = FuncInfo->PHINodesToUpdate.size(); pi != pe; ++pi) { - MachineInstr *PHI = FuncInfo->PHINodesToUpdate[pi].first; + MachineInstrBuilder PHI(*MF, FuncInfo->PHINodesToUpdate[pi].first); MachineBasicBlock *PHIBB = PHI->getParent(); assert(PHI->isPHI() && "This is not a machine PHI node that we are updating!"); // This is "default" BB. We have two jumps to it. From "header" BB and // from last "case" BB. - if (PHIBB == SDB->BitTestCases[i].Default) { - PHI->addOperand(MachineOperand:: - CreateReg(FuncInfo->PHINodesToUpdate[pi].second, - false)); - PHI->addOperand(MachineOperand::CreateMBB(SDB->BitTestCases[i].Parent)); - PHI->addOperand(MachineOperand:: - CreateReg(FuncInfo->PHINodesToUpdate[pi].second, - false)); - PHI->addOperand(MachineOperand::CreateMBB(SDB->BitTestCases[i].Cases. - back().ThisBB)); - } + if (PHIBB == SDB->BitTestCases[i].Default) + PHI.addReg(FuncInfo->PHINodesToUpdate[pi].second) + .addMBB(SDB->BitTestCases[i].Parent) + .addReg(FuncInfo->PHINodesToUpdate[pi].second) + .addMBB(SDB->BitTestCases[i].Cases.back().ThisBB); // One of "cases" BB. for (unsigned j = 0, ej = SDB->BitTestCases[i].Cases.size(); j != ej; ++j) { MachineBasicBlock* cBB = SDB->BitTestCases[i].Cases[j].ThisBB; - if (cBB->isSuccessor(PHIBB)) { - PHI->addOperand(MachineOperand:: - CreateReg(FuncInfo->PHINodesToUpdate[pi].second, - false)); - PHI->addOperand(MachineOperand::CreateMBB(cBB)); - } + if (cBB->isSuccessor(PHIBB)) + PHI.addReg(FuncInfo->PHINodesToUpdate[pi].second).addMBB(cBB); } } } @@ -1126,42 +1558,21 @@ SelectionDAGISel::FinishBasicBlock() { // Update PHI Nodes for (unsigned pi = 0, pe = FuncInfo->PHINodesToUpdate.size(); pi != pe; ++pi) { - MachineInstr *PHI = FuncInfo->PHINodesToUpdate[pi].first; + MachineInstrBuilder PHI(*MF, FuncInfo->PHINodesToUpdate[pi].first); MachineBasicBlock *PHIBB = PHI->getParent(); assert(PHI->isPHI() && "This is not a machine PHI node that we are updating!"); // "default" BB. We can go there only from header BB. - if (PHIBB == SDB->JTCases[i].second.Default) { - PHI->addOperand - (MachineOperand::CreateReg(FuncInfo->PHINodesToUpdate[pi].second, - false)); - PHI->addOperand - (MachineOperand::CreateMBB(SDB->JTCases[i].first.HeaderBB)); - } + if (PHIBB == SDB->JTCases[i].second.Default) + PHI.addReg(FuncInfo->PHINodesToUpdate[pi].second) + .addMBB(SDB->JTCases[i].first.HeaderBB); // JT BB. Just iterate over successors here - if (FuncInfo->MBB->isSuccessor(PHIBB)) { - PHI->addOperand - (MachineOperand::CreateReg(FuncInfo->PHINodesToUpdate[pi].second, - false)); - PHI->addOperand(MachineOperand::CreateMBB(FuncInfo->MBB)); - } + if (FuncInfo->MBB->isSuccessor(PHIBB)) + PHI.addReg(FuncInfo->PHINodesToUpdate[pi].second).addMBB(FuncInfo->MBB); } } SDB->JTCases.clear(); - // If the switch block involved a branch to one of the actual successors, we - // need to update PHI nodes in that block. - for (unsigned i = 0, e = FuncInfo->PHINodesToUpdate.size(); i != e; ++i) { - MachineInstr *PHI = FuncInfo->PHINodesToUpdate[i].first; - assert(PHI->isPHI() && - "This is not a machine PHI node that we are updating!"); - if (FuncInfo->MBB->isSuccessor(PHI->getParent())) { - PHI->addOperand( - MachineOperand::CreateReg(FuncInfo->PHINodesToUpdate[i].second, false)); - PHI->addOperand(MachineOperand::CreateMBB(FuncInfo->MBB)); - } - } - // If we generated any switch lowering information, build and codegen any // additional DAGs necessary. for (unsigned i = 0, e = SDB->SwitchCases.size(); i != e; ++i) { @@ -1195,18 +1606,16 @@ SelectionDAGISel::FinishBasicBlock() { // FuncInfo->MBB may have been removed from the CFG if a branch was // constant folded. if (ThisBB->isSuccessor(FuncInfo->MBB)) { - for (MachineBasicBlock::iterator Phi = FuncInfo->MBB->begin(); - Phi != FuncInfo->MBB->end() && Phi->isPHI(); - ++Phi) { + for (MachineBasicBlock::iterator + MBBI = FuncInfo->MBB->begin(), MBBE = FuncInfo->MBB->end(); + MBBI != MBBE && MBBI->isPHI(); ++MBBI) { + MachineInstrBuilder PHI(*MF, MBBI); // This value for this PHI node is recorded in PHINodesToUpdate. for (unsigned pn = 0; ; ++pn) { assert(pn != FuncInfo->PHINodesToUpdate.size() && "Didn't find PHI entry!"); - if (FuncInfo->PHINodesToUpdate[pn].first == Phi) { - Phi->addOperand(MachineOperand:: - CreateReg(FuncInfo->PHINodesToUpdate[pn].second, - false)); - Phi->addOperand(MachineOperand::CreateMBB(ThisBB)); + if (FuncInfo->PHINodesToUpdate[pn].first == PHI) { + PHI.addReg(FuncInfo->PHINodesToUpdate[pn].second).addMBB(ThisBB); break; } } @@ -1289,7 +1698,7 @@ bool SelectionDAGISel::CheckOrMask(SDValue LHS, ConstantSDNode *RHS, APInt NeededMask = DesiredMask & ~ActualMask; APInt KnownZero, KnownOne; - CurDAG->ComputeMaskedBits(LHS, NeededMask, KnownZero, KnownOne); + CurDAG->computeKnownBits(LHS, KnownZero, KnownOne); // If all the missing bits in the or are already known to be set, match! if ((NeededMask & KnownOne) == NeededMask) @@ -1301,11 +1710,10 @@ bool SelectionDAGISel::CheckOrMask(SDValue LHS, ConstantSDNode *RHS, return false; } - /// SelectInlineAsmMemoryOperands - Calls to this are automatically generated /// by tblgen. Others should not call it. void SelectionDAGISel:: -SelectInlineAsmMemoryOperands(std::vector &Ops) { +SelectInlineAsmMemoryOperands(std::vector &Ops, SDLoc DL) { std::vector InOps; std::swap(InOps, Ops); @@ -1328,16 +1736,30 @@ SelectInlineAsmMemoryOperands(std::vector &Ops) { } else { assert(InlineAsm::getNumOperandRegisters(Flags) == 1 && "Memory operand with multiple values?"); + + unsigned TiedToOperand; + if (InlineAsm::isUseOperandTiedToDef(Flags, TiedToOperand)) { + // We need the constraint ID from the operand this is tied to. + unsigned CurOp = InlineAsm::Op_FirstOperand; + Flags = cast(InOps[CurOp])->getZExtValue(); + for (; TiedToOperand; --TiedToOperand) { + CurOp += InlineAsm::getNumOperandRegisters(Flags)+1; + Flags = cast(InOps[CurOp])->getZExtValue(); + } + } + // Otherwise, this is a memory operand. Ask the target to select it. std::vector SelOps; - if (SelectInlineAsmMemoryOperand(InOps[i+1], 'm', SelOps)) + if (SelectInlineAsmMemoryOperand(InOps[i+1], + InlineAsm::getMemoryConstraintID(Flags), + SelOps)) report_fatal_error("Could not match memory address. Inline asm" " failure!"); // Add this to the output node. unsigned NewFlags = InlineAsm::getFlagWord(InlineAsm::Kind_Mem, SelOps.size()); - Ops.push_back(CurDAG->getTargetConstant(NewFlags, MVT::i32)); + Ops.push_back(CurDAG->getTargetConstant(NewFlags, DL, MVT::i32)); Ops.insert(Ops.end(), SelOps.begin(), SelOps.end()); i += 2; } @@ -1358,14 +1780,14 @@ static SDNode *findGlueUse(SDNode *N) { if (Use.getResNo() == FlagResNo) return Use.getUser(); } - return NULL; + return nullptr; } /// findNonImmUse - Return true if "Use" is a non-immediate use of "Def". /// This function recursively traverses up the operand chain, ignoring /// certain nodes. static bool findNonImmUse(SDNode *Use, SDNode* Def, SDNode *ImmedUse, - SDNode *Root, SmallPtrSet &Visited, + SDNode *Root, SmallPtrSetImpl &Visited, bool IgnoreChains) { // The NodeID's are given uniques ID's where a node ID is guaranteed to be // greater than all of its (recursive) operands. If we scan to a point where @@ -1380,7 +1802,7 @@ static bool findNonImmUse(SDNode *Use, SDNode* Def, SDNode *ImmedUse, // Don't revisit nodes if we already scanned it and didn't fail, we know we // won't fail if we scan it again. - if (!Visited.insert(Use)) + if (!Visited.insert(Use).second) return false; for (unsigned i = 0, e = Use->getNumOperands(); i != e; ++i) { @@ -1465,7 +1887,7 @@ bool SelectionDAGISel::IsLegalToFold(SDValue N, SDNode *U, SDNode *Root, EVT VT = Root->getValueType(Root->getNumValues()-1); while (VT == MVT::Glue) { SDNode *GU = findGlueUse(Root); - if (GU == NULL) + if (!GU) break; Root = GU; VT = Root->getValueType(Root->getNumValues()-1); @@ -1483,18 +1905,45 @@ bool SelectionDAGISel::IsLegalToFold(SDValue N, SDNode *U, SDNode *Root, } SDNode *SelectionDAGISel::Select_INLINEASM(SDNode *N) { + SDLoc DL(N); + std::vector Ops(N->op_begin(), N->op_end()); - SelectInlineAsmMemoryOperands(Ops); + SelectInlineAsmMemoryOperands(Ops, DL); - std::vector VTs; - VTs.push_back(MVT::Other); - VTs.push_back(MVT::Glue); - SDValue New = CurDAG->getNode(ISD::INLINEASM, N->getDebugLoc(), - VTs, &Ops[0], Ops.size()); + const EVT VTs[] = {MVT::Other, MVT::Glue}; + SDValue New = CurDAG->getNode(ISD::INLINEASM, DL, VTs, Ops); New->setNodeId(-1); return New.getNode(); } +SDNode +*SelectionDAGISel::Select_READ_REGISTER(SDNode *Op) { + SDLoc dl(Op); + MDNodeSDNode *MD = dyn_cast(Op->getOperand(0)); + const MDString *RegStr = dyn_cast(MD->getMD()->getOperand(0)); + unsigned Reg = + TLI->getRegisterByName(RegStr->getString().data(), Op->getValueType(0)); + SDValue New = CurDAG->getCopyFromReg( + CurDAG->getEntryNode(), dl, Reg, Op->getValueType(0)); + New->setNodeId(-1); + return New.getNode(); +} + +SDNode +*SelectionDAGISel::Select_WRITE_REGISTER(SDNode *Op) { + SDLoc dl(Op); + MDNodeSDNode *MD = dyn_cast(Op->getOperand(1)); + const MDString *RegStr = dyn_cast(MD->getMD()->getOperand(0)); + unsigned Reg = TLI->getRegisterByName(RegStr->getString().data(), + Op->getOperand(2).getValueType()); + SDValue New = CurDAG->getCopyToReg( + CurDAG->getEntryNode(), dl, Reg, Op->getOperand(2)); + New->setNodeId(-1); + return New.getNode(); +} + + + SDNode *SelectionDAGISel::Select_UNDEF(SDNode *N) { return CurDAG->SelectNodeTo(N, TargetOpcode::IMPLICIT_DEF,N->getValueType(0)); } @@ -1527,12 +1976,10 @@ UpdateChainsAndGlue(SDNode *NodeToMatch, SDValue InputChain, bool isMorphNodeTo) { SmallVector NowDeadNodes; - ISelUpdater ISU(ISelPosition); - // Now that all the normal results are replaced, we replace the chain and // glue results if present. if (!ChainNodesMatched.empty()) { - assert(InputChain.getNode() != 0 && + assert(InputChain.getNode() && "Matched input chains but didn't produce a chain"); // Loop over all of the nodes we matched that produced a chain result. // Replace all the chain results with the final chain we ended up with. @@ -1552,7 +1999,7 @@ UpdateChainsAndGlue(SDNode *NodeToMatch, SDValue InputChain, if (ChainVal.getValueType() == MVT::Glue) ChainVal = ChainVal.getValue(ChainVal->getNumValues()-2); assert(ChainVal.getValueType() == MVT::Other && "Not a chain?"); - CurDAG->ReplaceAllUsesOfValueWith(ChainVal, InputChain, &ISU); + CurDAG->ReplaceAllUsesOfValueWith(ChainVal, InputChain); // If the node became dead and we haven't already seen it, delete it. if (ChainNode->use_empty() && @@ -1563,7 +2010,7 @@ UpdateChainsAndGlue(SDNode *NodeToMatch, SDValue InputChain, // If the result produces glue, update any glue results in the matched // pattern with the glue result. - if (InputGlue.getNode() != 0) { + if (InputGlue.getNode()) { // Handle any interior nodes explicitly marked. for (unsigned i = 0, e = GlueResultNodesMatched.size(); i != e; ++i) { SDNode *FRN = GlueResultNodesMatched[i]; @@ -1575,7 +2022,7 @@ UpdateChainsAndGlue(SDNode *NodeToMatch, SDValue InputChain, assert(FRN->getValueType(FRN->getNumValues()-1) == MVT::Glue && "Doesn't have a glue result"); CurDAG->ReplaceAllUsesOfValueWith(SDValue(FRN, FRN->getNumValues()-1), - InputGlue, &ISU); + InputGlue); // If the node became dead and we haven't already seen it, delete it. if (FRN->use_empty() && @@ -1585,9 +2032,9 @@ UpdateChainsAndGlue(SDNode *NodeToMatch, SDValue InputChain, } if (!NowDeadNodes.empty()) - CurDAG->RemoveDeadNodes(NowDeadNodes, &ISU); + CurDAG->RemoveDeadNodes(NowDeadNodes); - DEBUG(errs() << "ISEL: Match complete!\n"); + DEBUG(dbgs() << "ISEL: Match complete!\n"); } enum ChainResult { @@ -1606,7 +2053,7 @@ enum ChainResult { /// The walk we do here is guaranteed to be small because we quickly get down to /// already selected nodes "below" us. static ChainResult -WalkChainUsers(SDNode *ChainedNode, +WalkChainUsers(const SDNode *ChainedNode, SmallVectorImpl &ChainedNodesInPattern, SmallVectorImpl &InteriorChainedNodes) { ChainResult Result = CR_Simple; @@ -1618,17 +2065,20 @@ WalkChainUsers(SDNode *ChainedNode, SDNode *User = *UI; + if (User->getOpcode() == ISD::HANDLENODE) // Root of the graph. + continue; + // If we see an already-selected machine node, then we've gone beyond the // pattern that we're selecting down into the already selected chunk of the // DAG. + unsigned UserOpcode = User->getOpcode(); if (User->isMachineOpcode() || - User->getOpcode() == ISD::HANDLENODE) // Root of the graph. - continue; - - if (User->getOpcode() == ISD::CopyToReg || - User->getOpcode() == ISD::CopyFromReg || - User->getOpcode() == ISD::INLINEASM || - User->getOpcode() == ISD::EH_LABEL) { + UserOpcode == ISD::CopyToReg || + UserOpcode == ISD::CopyFromReg || + UserOpcode == ISD::INLINEASM || + UserOpcode == ISD::EH_LABEL || + UserOpcode == ISD::LIFETIME_START || + UserOpcode == ISD::LIFETIME_END) { // If their node ID got reset to -1 then they've already been selected. // Treat them like a MachineOpcode. if (User->getNodeId() == -1) @@ -1760,17 +2210,16 @@ HandleMergeInputChains(SmallVectorImpl &ChainNodesMatched, } } - SDValue Res; if (InputChains.size() == 1) return InputChains[0]; - return CurDAG->getNode(ISD::TokenFactor, ChainNodesMatched[0]->getDebugLoc(), - MVT::Other, &InputChains[0], InputChains.size()); + return CurDAG->getNode(ISD::TokenFactor, SDLoc(ChainNodesMatched[0]), + MVT::Other, InputChains); } /// MorphNode - Handle morphing a node in place for the selector. SDNode *SelectionDAGISel:: MorphNode(SDNode *Node, unsigned TargetOpc, SDVTList VTList, - const SDValue *Ops, unsigned NumOps, unsigned EmitNodeInfo) { + ArrayRef Ops, unsigned EmitNodeInfo) { // It is possible we're using MorphNodeTo to replace a node with no // normal results with one that has a normal result (or we could be // adding a chain) and the input could have glue and chains as well. @@ -1790,7 +2239,7 @@ MorphNode(SDNode *Node, unsigned TargetOpc, SDVTList VTList, // Call the underlying SelectionDAG routine to do the transmogrification. Note // that this deletes operands of the old node that become dead. - SDNode *Res = CurDAG->MorphNodeTo(Node, ~TargetOpc, VTList, Ops, NumOps); + SDNode *Res = CurDAG->MorphNodeTo(Node, ~TargetOpc, VTList, Ops); // MorphNodeTo can operate in two ways: if an existing node with the // specified operands exists, it can just return it. Otherwise, it @@ -1825,7 +2274,7 @@ MorphNode(SDNode *Node, unsigned TargetOpc, SDVTList VTList, return Res; } -/// CheckPatternPredicate - Implements OP_CheckPatternPredicate. +/// CheckSame - Implements OP_CheckSame. LLVM_ATTRIBUTE_ALWAYS_INLINE static bool CheckSame(const unsigned char *MatcherTable, unsigned &MatcherIndex, SDValue N, @@ -1836,17 +2285,29 @@ CheckSame(const unsigned char *MatcherTable, unsigned &MatcherIndex, return N == RecordedNodes[RecNo].first; } +/// CheckChildSame - Implements OP_CheckChildXSame. +LLVM_ATTRIBUTE_ALWAYS_INLINE static bool +CheckChildSame(const unsigned char *MatcherTable, unsigned &MatcherIndex, + SDValue N, + const SmallVectorImpl > &RecordedNodes, + unsigned ChildNo) { + if (ChildNo >= N.getNumOperands()) + return false; // Match fails if out of range child #. + return ::CheckSame(MatcherTable, MatcherIndex, N.getOperand(ChildNo), + RecordedNodes); +} + /// CheckPatternPredicate - Implements OP_CheckPatternPredicate. LLVM_ATTRIBUTE_ALWAYS_INLINE static bool CheckPatternPredicate(const unsigned char *MatcherTable, unsigned &MatcherIndex, - SelectionDAGISel &SDISel) { + const SelectionDAGISel &SDISel) { return SDISel.CheckPatternPredicate(MatcherTable[MatcherIndex++]); } /// CheckNodePredicate - Implements OP_CheckNodePredicate. LLVM_ATTRIBUTE_ALWAYS_INLINE static bool CheckNodePredicate(const unsigned char *MatcherTable, unsigned &MatcherIndex, - SelectionDAGISel &SDISel, SDNode *N) { + const SelectionDAGISel &SDISel, SDNode *N) { return SDISel.CheckNodePredicate(N, MatcherTable[MatcherIndex++]); } @@ -1860,24 +2321,22 @@ CheckOpcode(const unsigned char *MatcherTable, unsigned &MatcherIndex, LLVM_ATTRIBUTE_ALWAYS_INLINE static bool CheckType(const unsigned char *MatcherTable, unsigned &MatcherIndex, - SDValue N, const TargetLowering &TLI) { + SDValue N, const TargetLowering *TLI) { MVT::SimpleValueType VT = (MVT::SimpleValueType)MatcherTable[MatcherIndex++]; if (N.getValueType() == VT) return true; // Handle the case when VT is iPTR. - return VT == MVT::iPTR && N.getValueType() == TLI.getPointerTy(); + return VT == MVT::iPTR && N.getValueType() == TLI->getPointerTy(); } LLVM_ATTRIBUTE_ALWAYS_INLINE static bool CheckChildType(const unsigned char *MatcherTable, unsigned &MatcherIndex, - SDValue N, const TargetLowering &TLI, - unsigned ChildNo) { + SDValue N, const TargetLowering *TLI, unsigned ChildNo) { if (ChildNo >= N.getNumOperands()) return false; // Match fails if out of range child #. return ::CheckType(MatcherTable, MatcherIndex, N.getOperand(ChildNo), TLI); } - LLVM_ATTRIBUTE_ALWAYS_INLINE static bool CheckCondCode(const unsigned char *MatcherTable, unsigned &MatcherIndex, SDValue N) { @@ -1887,13 +2346,13 @@ CheckCondCode(const unsigned char *MatcherTable, unsigned &MatcherIndex, LLVM_ATTRIBUTE_ALWAYS_INLINE static bool CheckValueType(const unsigned char *MatcherTable, unsigned &MatcherIndex, - SDValue N, const TargetLowering &TLI) { + SDValue N, const TargetLowering *TLI) { MVT::SimpleValueType VT = (MVT::SimpleValueType)MatcherTable[MatcherIndex++]; if (cast(N)->getVT() == VT) return true; // Handle the case when VT is iPTR. - return VT == MVT::iPTR && cast(N)->getVT() == TLI.getPointerTy(); + return VT == MVT::iPTR && cast(N)->getVT() == TLI->getPointerTy(); } LLVM_ATTRIBUTE_ALWAYS_INLINE static bool @@ -1904,12 +2363,20 @@ CheckInteger(const unsigned char *MatcherTable, unsigned &MatcherIndex, Val = GetVBR(Val, MatcherTable, MatcherIndex); ConstantSDNode *C = dyn_cast(N); - return C != 0 && C->getSExtValue() == Val; + return C && C->getSExtValue() == Val; +} + +LLVM_ATTRIBUTE_ALWAYS_INLINE static bool +CheckChildInteger(const unsigned char *MatcherTable, unsigned &MatcherIndex, + SDValue N, unsigned ChildNo) { + if (ChildNo >= N.getNumOperands()) + return false; // Match fails if out of range child #. + return ::CheckInteger(MatcherTable, MatcherIndex, N.getOperand(ChildNo)); } LLVM_ATTRIBUTE_ALWAYS_INLINE static bool CheckAndImm(const unsigned char *MatcherTable, unsigned &MatcherIndex, - SDValue N, SelectionDAGISel &SDISel) { + SDValue N, const SelectionDAGISel &SDISel) { int64_t Val = MatcherTable[MatcherIndex++]; if (Val & 128) Val = GetVBR(Val, MatcherTable, MatcherIndex); @@ -1917,12 +2384,12 @@ CheckAndImm(const unsigned char *MatcherTable, unsigned &MatcherIndex, if (N->getOpcode() != ISD::AND) return false; ConstantSDNode *C = dyn_cast(N->getOperand(1)); - return C != 0 && SDISel.CheckAndMask(N.getOperand(0), C, Val); + return C && SDISel.CheckAndMask(N.getOperand(0), C, Val); } LLVM_ATTRIBUTE_ALWAYS_INLINE static bool CheckOrImm(const unsigned char *MatcherTable, unsigned &MatcherIndex, - SDValue N, SelectionDAGISel &SDISel) { + SDValue N, const SelectionDAGISel &SDISel) { int64_t Val = MatcherTable[MatcherIndex++]; if (Val & 128) Val = GetVBR(Val, MatcherTable, MatcherIndex); @@ -1930,7 +2397,7 @@ CheckOrImm(const unsigned char *MatcherTable, unsigned &MatcherIndex, if (N->getOpcode() != ISD::OR) return false; ConstantSDNode *C = dyn_cast(N->getOperand(1)); - return C != 0 && SDISel.CheckOrMask(N.getOperand(0), C, Val); + return C && SDISel.CheckOrMask(N.getOperand(0), C, Val); } /// IsPredicateKnownToFail - If we know how and can do so without pushing a @@ -1941,7 +2408,8 @@ CheckOrImm(const unsigned char *MatcherTable, unsigned &MatcherIndex, /// MatcherIndex to continue with. static unsigned IsPredicateKnownToFail(const unsigned char *Table, unsigned Index, SDValue N, - bool &Result, SelectionDAGISel &SDISel, + bool &Result, + const SelectionDAGISel &SDISel, SmallVectorImpl > &RecordedNodes) { switch (Table[Index++]) { default: @@ -1950,6 +2418,13 @@ static unsigned IsPredicateKnownToFail(const unsigned char *Table, case SelectionDAGISel::OPC_CheckSame: Result = !::CheckSame(Table, Index, N, RecordedNodes); return Index; + case SelectionDAGISel::OPC_CheckChild0Same: + case SelectionDAGISel::OPC_CheckChild1Same: + case SelectionDAGISel::OPC_CheckChild2Same: + case SelectionDAGISel::OPC_CheckChild3Same: + Result = !::CheckChildSame(Table, Index, N, RecordedNodes, + Table[Index-1] - SelectionDAGISel::OPC_CheckChild0Same); + return Index; case SelectionDAGISel::OPC_CheckPatternPredicate: Result = !::CheckPatternPredicate(Table, Index, SDISel); return Index; @@ -1971,7 +2446,8 @@ static unsigned IsPredicateKnownToFail(const unsigned char *Table, case SelectionDAGISel::OPC_CheckChild6Type: case SelectionDAGISel::OPC_CheckChild7Type: Result = !::CheckChildType(Table, Index, N, SDISel.TLI, - Table[Index-1] - SelectionDAGISel::OPC_CheckChild0Type); + Table[Index - 1] - + SelectionDAGISel::OPC_CheckChild0Type); return Index; case SelectionDAGISel::OPC_CheckCondCode: Result = !::CheckCondCode(Table, Index, N); @@ -1982,6 +2458,14 @@ static unsigned IsPredicateKnownToFail(const unsigned char *Table, case SelectionDAGISel::OPC_CheckInteger: Result = !::CheckInteger(Table, Index, N); return Index; + case SelectionDAGISel::OPC_CheckChild0Integer: + case SelectionDAGISel::OPC_CheckChild1Integer: + case SelectionDAGISel::OPC_CheckChild2Integer: + case SelectionDAGISel::OPC_CheckChild3Integer: + case SelectionDAGISel::OPC_CheckChild4Integer: + Result = !::CheckChildInteger(Table, Index, N, + Table[Index-1] - SelectionDAGISel::OPC_CheckChild0Integer); + return Index; case SelectionDAGISel::OPC_CheckAndImm: Result = !::CheckAndImm(Table, Index, N, SDISel); return Index; @@ -2013,6 +2497,42 @@ struct MatchScope { bool HasChainNodesMatched, HasGlueResultNodesMatched; }; +/// \\brief A DAG update listener to keep the matching state +/// (i.e. RecordedNodes and MatchScope) uptodate if the target is allowed to +/// change the DAG while matching. X86 addressing mode matcher is an example +/// for this. +class MatchStateUpdater : public SelectionDAG::DAGUpdateListener +{ + SmallVectorImpl > &RecordedNodes; + SmallVectorImpl &MatchScopes; +public: + MatchStateUpdater(SelectionDAG &DAG, + SmallVectorImpl > &RN, + SmallVectorImpl &MS) : + SelectionDAG::DAGUpdateListener(DAG), + RecordedNodes(RN), MatchScopes(MS) { } + + void NodeDeleted(SDNode *N, SDNode *E) override { + // Some early-returns here to avoid the search if we deleted the node or + // if the update comes from MorphNodeTo (MorphNodeTo is the last thing we + // do, so it's unnecessary to update matching state at that point). + // Neither of these can occur currently because we only install this + // update listener during matching a complex patterns. + if (!E || E->isMachineOpcode()) + return; + // Performing linear search here does not matter because we almost never + // run this code. You'd have to have a CSE during complex pattern + // matching. + for (auto &I : RecordedNodes) + if (I.first.getNode() == N) + I.first.setNode(E); + + for (auto &I : MatchScopes) + for (auto &J : I.NodeStack) + if (J.getNode() == N) + J.setNode(E); + } +}; } SDNode *SelectionDAGISel:: @@ -2025,8 +2545,7 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, case ISD::EntryToken: // These nodes remain the same. case ISD::BasicBlock: case ISD::Register: - //case ISD::VALUETYPE: - //case ISD::CONDCODE: + case ISD::RegisterMask: case ISD::HANDLENODE: case ISD::MDNODE_SDNODE: case ISD::TargetConstant: @@ -2042,14 +2561,18 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, case ISD::CopyFromReg: case ISD::CopyToReg: case ISD::EH_LABEL: + case ISD::LIFETIME_START: + case ISD::LIFETIME_END: NodeToMatch->setNodeId(-1); // Mark selected. - return 0; + return nullptr; case ISD::AssertSext: case ISD::AssertZext: CurDAG->ReplaceAllUsesOfValueWith(SDValue(NodeToMatch, 0), NodeToMatch->getOperand(0)); - return 0; + return nullptr; case ISD::INLINEASM: return Select_INLINEASM(NodeToMatch); + case ISD::READ_REGISTER: return Select_READ_REGISTER(NodeToMatch); + case ISD::WRITE_REGISTER: return Select_WRITE_REGISTER(NodeToMatch); case ISD::UNDEF: return Select_UNDEF(NodeToMatch); } @@ -2085,9 +2608,9 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, SmallVector ChainNodesMatched; SmallVector GlueResultNodesMatched; - DEBUG(errs() << "ISEL: Starting pattern match on root node: "; + DEBUG(dbgs() << "ISEL: Starting pattern match on root node: "; NodeToMatch->dump(CurDAG); - errs() << '\n'); + dbgs() << '\n'); // Determine where to start the interpreter. Normally we start at opcode #0, // but if the state machine starts with an OPC_SwitchOpcode, then we @@ -2099,7 +2622,7 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, // Already computed the OpcodeOffset table, just index into it. if (N.getOpcode() < OpcodeOffset.size()) MatcherIndex = OpcodeOffset[N.getOpcode()]; - DEBUG(errs() << " Initial Opcode index to " << MatcherIndex << "\n"); + DEBUG(dbgs() << " Initial Opcode index to " << MatcherIndex << "\n"); } else if (MatcherTable[0] == OPC_SwitchOpcode) { // Otherwise, the table isn't computed, but the state machine does start @@ -2166,7 +2689,7 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, if (!Result) break; - DEBUG(errs() << " Skipped scope entry (due to false predicate) at " + DEBUG(dbgs() << " Skipped scope entry (due to false predicate) at " << "index " << MatcherIndexOfPredicate << ", continuing at " << FailIndex << "\n"); ++NumDAGIselRetries; @@ -2195,7 +2718,7 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, } case OPC_RecordNode: { // Remember this node, it may end up being an operand in the pattern. - SDNode *Parent = 0; + SDNode *Parent = nullptr; if (NodeStack.size() > 1) Parent = NodeStack[NodeStack.size()-2].getNode(); RecordedNodes.push_back(std::make_pair(N, Parent)); @@ -2244,6 +2767,14 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, case OPC_CheckSame: if (!::CheckSame(MatcherTable, MatcherIndex, N, RecordedNodes)) break; continue; + + case OPC_CheckChild0Same: case OPC_CheckChild1Same: + case OPC_CheckChild2Same: case OPC_CheckChild3Same: + if (!::CheckChildSame(MatcherTable, MatcherIndex, N, RecordedNodes, + Opcode-OPC_CheckChild0Same)) + break; + continue; + case OPC_CheckPatternPredicate: if (!::CheckPatternPredicate(MatcherTable, MatcherIndex, *this)) break; continue; @@ -2256,6 +2787,14 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, unsigned CPNum = MatcherTable[MatcherIndex++]; unsigned RecNo = MatcherTable[MatcherIndex++]; assert(RecNo < RecordedNodes.size() && "Invalid CheckComplexPat"); + + // If target can modify DAG during matching, keep the matching state + // consistent. + std::unique_ptr MSU; + if (ComplexPatternFuncMutatesDAG()) + MSU.reset(new MatchStateUpdater(*CurDAG, RecordedNodes, + MatchScopes)); + if (!CheckComplexPattern(NodeToMatch, RecordedNodes[RecNo].second, RecordedNodes[RecNo].first, CPNum, RecordedNodes)) @@ -2267,7 +2806,8 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, continue; case OPC_CheckType: - if (!::CheckType(MatcherTable, MatcherIndex, N, TLI)) break; + if (!::CheckType(MatcherTable, MatcherIndex, N, TLI)) + break; continue; case OPC_SwitchOpcode: { @@ -2296,13 +2836,13 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, if (CaseSize == 0) break; // Otherwise, execute the case we found. - DEBUG(errs() << " OpcodeSwitch from " << SwitchStart + DEBUG(dbgs() << " OpcodeSwitch from " << SwitchStart << " to " << MatcherIndex << "\n"); continue; } case OPC_SwitchType: { - MVT CurNodeVT = N.getValueType().getSimpleVT(); + MVT CurNodeVT = N.getSimpleValueType(); unsigned SwitchStart = MatcherIndex-1; (void)SwitchStart; unsigned CaseSize; while (1) { @@ -2314,7 +2854,7 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, MVT CaseVT = (MVT::SimpleValueType)MatcherTable[MatcherIndex++]; if (CaseVT == MVT::iPTR) - CaseVT = TLI.getPointerTy(); + CaseVT = TLI->getPointerTy(); // If the VT matches, then we will execute this case. if (CurNodeVT == CaseVT) @@ -2328,7 +2868,7 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, if (CaseSize == 0) break; // Otherwise, execute the case we found. - DEBUG(errs() << " TypeSwitch[" << EVT(CurNodeVT).getEVTString() + DEBUG(dbgs() << " TypeSwitch[" << EVT(CurNodeVT).getEVTString() << "] from " << SwitchStart << " to " << MatcherIndex<<'\n'); continue; } @@ -2344,11 +2884,18 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, if (!::CheckCondCode(MatcherTable, MatcherIndex, N)) break; continue; case OPC_CheckValueType: - if (!::CheckValueType(MatcherTable, MatcherIndex, N, TLI)) break; + if (!::CheckValueType(MatcherTable, MatcherIndex, N, TLI)) + break; continue; case OPC_CheckInteger: if (!::CheckInteger(MatcherTable, MatcherIndex, N)) break; continue; + case OPC_CheckChild0Integer: case OPC_CheckChild1Integer: + case OPC_CheckChild2Integer: case OPC_CheckChild3Integer: + case OPC_CheckChild4Integer: + if (!::CheckChildInteger(MatcherTable, MatcherIndex, N, + Opcode-OPC_CheckChild0Integer)) break; + continue; case OPC_CheckAndImm: if (!::CheckAndImm(MatcherTable, MatcherIndex, N, *this)) break; continue; @@ -2386,7 +2933,8 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, if (Val & 128) Val = GetVBR(Val, MatcherTable, MatcherIndex); RecordedNodes.push_back(std::pair( - CurDAG->getTargetConstant(Val, VT), (SDNode*)0)); + CurDAG->getTargetConstant(Val, SDLoc(NodeToMatch), + VT), nullptr)); continue; } case OPC_EmitRegister: { @@ -2394,7 +2942,7 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, (MVT::SimpleValueType)MatcherTable[MatcherIndex++]; unsigned RegNo = MatcherTable[MatcherIndex++]; RecordedNodes.push_back(std::pair( - CurDAG->getRegister(RegNo, VT), (SDNode*)0)); + CurDAG->getRegister(RegNo, VT), nullptr)); continue; } case OPC_EmitRegister2: { @@ -2406,22 +2954,24 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, unsigned RegNo = MatcherTable[MatcherIndex++]; RegNo |= MatcherTable[MatcherIndex++] << 8; RecordedNodes.push_back(std::pair( - CurDAG->getRegister(RegNo, VT), (SDNode*)0)); + CurDAG->getRegister(RegNo, VT), nullptr)); continue; } case OPC_EmitConvertToTarget: { // Convert from IMM/FPIMM to target version. unsigned RecNo = MatcherTable[MatcherIndex++]; - assert(RecNo < RecordedNodes.size() && "Invalid CheckSame"); + assert(RecNo < RecordedNodes.size() && "Invalid EmitConvertToTarget"); SDValue Imm = RecordedNodes[RecNo].first; if (Imm->getOpcode() == ISD::Constant) { - int64_t Val = cast(Imm)->getZExtValue(); - Imm = CurDAG->getTargetConstant(Val, Imm.getValueType()); + const ConstantInt *Val=cast(Imm)->getConstantIntValue(); + Imm = CurDAG->getConstant(*Val, SDLoc(NodeToMatch), Imm.getValueType(), + true); } else if (Imm->getOpcode() == ISD::ConstantFP) { const ConstantFP *Val=cast(Imm)->getConstantFPValue(); - Imm = CurDAG->getTargetConstantFP(*Val, Imm.getValueType()); + Imm = CurDAG->getConstantFP(*Val, SDLoc(NodeToMatch), + Imm.getValueType(), true); } RecordedNodes.push_back(std::make_pair(Imm, RecordedNodes[RecNo].second)); @@ -2431,14 +2981,14 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, case OPC_EmitMergeInputChains1_0: // OPC_EmitMergeInputChains, 1, 0 case OPC_EmitMergeInputChains1_1: { // OPC_EmitMergeInputChains, 1, 1 // These are space-optimized forms of OPC_EmitMergeInputChains. - assert(InputChain.getNode() == 0 && + assert(!InputChain.getNode() && "EmitMergeInputChains should be the first chain producing node"); assert(ChainNodesMatched.empty() && "Should only have one EmitMergeInputChains per match"); // Read all of the chained nodes. unsigned RecNo = Opcode == OPC_EmitMergeInputChains1_1; - assert(RecNo < RecordedNodes.size() && "Invalid CheckSame"); + assert(RecNo < RecordedNodes.size() && "Invalid EmitMergeInputChains"); ChainNodesMatched.push_back(RecordedNodes[RecNo].first.getNode()); // FIXME: What if other value results of the node have uses not matched @@ -2452,13 +3002,13 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, // Merge the input chains if they are not intra-pattern references. InputChain = HandleMergeInputChains(ChainNodesMatched, CurDAG); - if (InputChain.getNode() == 0) + if (!InputChain.getNode()) break; // Failed to merge. continue; } case OPC_EmitMergeInputChains: { - assert(InputChain.getNode() == 0 && + assert(!InputChain.getNode() && "EmitMergeInputChains should be the first chain producing node"); // This node gets a list of nodes we matched in the input that have // chains. We want to token factor all of the input chains to these nodes @@ -2475,7 +3025,7 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, // Read all of the chained nodes. for (unsigned i = 0; i != NumChains; ++i) { unsigned RecNo = MatcherTable[MatcherIndex++]; - assert(RecNo < RecordedNodes.size() && "Invalid CheckSame"); + assert(RecNo < RecordedNodes.size() && "Invalid EmitMergeInputChains"); ChainNodesMatched.push_back(RecordedNodes[RecNo].first.getNode()); // FIXME: What if other value results of the node have uses not matched @@ -2494,7 +3044,7 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, // Merge the input chains if they are not intra-pattern references. InputChain = HandleMergeInputChains(ChainNodesMatched, CurDAG); - if (InputChain.getNode() == 0) + if (!InputChain.getNode()) break; // Failed to merge. continue; @@ -2502,13 +3052,13 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, case OPC_EmitCopyToReg: { unsigned RecNo = MatcherTable[MatcherIndex++]; - assert(RecNo < RecordedNodes.size() && "Invalid CheckSame"); + assert(RecNo < RecordedNodes.size() && "Invalid EmitCopyToReg"); unsigned DestPhysReg = MatcherTable[MatcherIndex++]; - if (InputChain.getNode() == 0) + if (!InputChain.getNode()) InputChain = CurDAG->getEntryNode(); - InputChain = CurDAG->getCopyToReg(InputChain, NodeToMatch->getDebugLoc(), + InputChain = CurDAG->getCopyToReg(InputChain, SDLoc(NodeToMatch), DestPhysReg, RecordedNodes[RecNo].first, InputGlue); @@ -2519,9 +3069,9 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, case OPC_EmitNodeXForm: { unsigned XFormNo = MatcherTable[MatcherIndex++]; unsigned RecNo = MatcherTable[MatcherIndex++]; - assert(RecNo < RecordedNodes.size() && "Invalid CheckSame"); + assert(RecNo < RecordedNodes.size() && "Invalid EmitNodeXForm"); SDValue Res = RunSDNodeXForm(RecordedNodes[RecNo].first, XFormNo); - RecordedNodes.push_back(std::pair(Res, (SDNode*) 0)); + RecordedNodes.push_back(std::pair(Res, nullptr)); continue; } @@ -2536,7 +3086,8 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, for (unsigned i = 0; i != NumVTs; ++i) { MVT::SimpleValueType VT = (MVT::SimpleValueType)MatcherTable[MatcherIndex++]; - if (VT == MVT::iPTR) VT = TLI.getPointerTy().SimpleTy; + if (VT == MVT::iPTR) + VT = TLI->getPointerTy().SimpleTy; VTs.push_back(VT); } @@ -2553,7 +3104,7 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, else if (VTs.size() == 2) VTList = CurDAG->getVTList(VTs[0], VTs[1]); else - VTList = CurDAG->getVTList(VTs.data(), VTs.size()); + VTList = CurDAG->getVTList(VTs); // Get the operand list. unsigned NumOps = MatcherTable[MatcherIndex++]; @@ -2587,27 +3138,31 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, // If this has chain/glue inputs, add them. if (EmitNodeInfo & OPFL_Chain) Ops.push_back(InputChain); - if ((EmitNodeInfo & OPFL_GlueInput) && InputGlue.getNode() != 0) + if ((EmitNodeInfo & OPFL_GlueInput) && InputGlue.getNode() != nullptr) Ops.push_back(InputGlue); // Create the node. - SDNode *Res = 0; + SDNode *Res = nullptr; if (Opcode != OPC_MorphNodeTo) { // If this is a normal EmitNode command, just create the new node and // add the results to the RecordedNodes list. - Res = CurDAG->getMachineNode(TargetOpc, NodeToMatch->getDebugLoc(), - VTList, Ops.data(), Ops.size()); + Res = CurDAG->getMachineNode(TargetOpc, SDLoc(NodeToMatch), + VTList, Ops); // Add all the non-glue/non-chain results to the RecordedNodes list. for (unsigned i = 0, e = VTs.size(); i != e; ++i) { if (VTs[i] == MVT::Other || VTs[i] == MVT::Glue) break; RecordedNodes.push_back(std::pair(SDValue(Res, i), - (SDNode*) 0)); + nullptr)); } + } else if (NodeToMatch->getOpcode() != ISD::DELETED_NODE) { + Res = MorphNode(NodeToMatch, TargetOpc, VTList, Ops, EmitNodeInfo); } else { - Res = MorphNode(NodeToMatch, TargetOpc, VTList, Ops.data(), Ops.size(), - EmitNodeInfo); + // NodeToMatch was eliminated by CSE when the target changed the DAG. + // We will visit the equivalent node later. + DEBUG(dbgs() << "Node was eliminated by CSE\n"); + return nullptr; } // If the node had chain/glue results, update our notion of the current @@ -2628,13 +3183,13 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, if (EmitNodeInfo & OPFL_MemRefs) { // Only attach load or store memory operands if the generated // instruction may load or store. - const MCInstrDesc &MCID = TM.getInstrInfo()->get(TargetOpc); + const MCInstrDesc &MCID = TII->get(TargetOpc); bool mayLoad = MCID.mayLoad(); bool mayStore = MCID.mayStore(); unsigned NumMemRefs = 0; - for (SmallVector::const_iterator I = - MatchedMemRefs.begin(), E = MatchedMemRefs.end(); I != E; ++I) { + for (SmallVectorImpl::const_iterator I = + MatchedMemRefs.begin(), E = MatchedMemRefs.end(); I != E; ++I) { if ((*I)->isLoad()) { if (mayLoad) ++NumMemRefs; @@ -2650,8 +3205,8 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, MF->allocateMemRefsArray(NumMemRefs); MachineSDNode::mmo_iterator MemRefsPos = MemRefs; - for (SmallVector::const_iterator I = - MatchedMemRefs.begin(), E = MatchedMemRefs.end(); I != E; ++I) { + for (SmallVectorImpl::const_iterator I = + MatchedMemRefs.begin(), E = MatchedMemRefs.end(); I != E; ++I) { if ((*I)->isLoad()) { if (mayLoad) *MemRefsPos++ = *I; @@ -2667,9 +3222,9 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, ->setMemRefs(MemRefs, MemRefs + NumMemRefs); } - DEBUG(errs() << " " + DEBUG(dbgs() << " " << (Opcode == OPC_MorphNodeTo ? "Morphed" : "Created") - << " node: "; Res->dump(CurDAG); errs() << "\n"); + << " node: "; Res->dump(CurDAG); dbgs() << "\n"); // If this was a MorphNodeTo then we're completely done! if (Opcode == OPC_MorphNodeTo) { @@ -2691,7 +3246,7 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, if (RecNo & 128) RecNo = GetVBR(RecNo, MatcherTable, MatcherIndex); - assert(RecNo < RecordedNodes.size() && "Invalid CheckSame"); + assert(RecNo < RecordedNodes.size() && "Invalid MarkGlueResults"); GlueResultNodesMatched.push_back(RecordedNodes[RecNo].first.getNode()); } continue; @@ -2708,7 +3263,7 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, if (ResSlot & 128) ResSlot = GetVBR(ResSlot, MatcherTable, MatcherIndex); - assert(ResSlot < RecordedNodes.size() && "Invalid CheckSame"); + assert(ResSlot < RecordedNodes.size() && "Invalid CompleteMatch"); SDValue Res = RecordedNodes[ResSlot].first; assert(i < NodeToMatch->getNumValues() && @@ -2737,19 +3292,19 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, // FIXME: We just return here, which interacts correctly with SelectRoot // above. We should fix this to not return an SDNode* anymore. - return 0; + return nullptr; } } // If the code reached this point, then the match failed. See if there is // another child to try in the current 'Scope', otherwise pop it until we // find a case to check. - DEBUG(errs() << " Match failed at index " << CurrentOpcodeIndex << "\n"); + DEBUG(dbgs() << " Match failed at index " << CurrentOpcodeIndex << "\n"); ++NumDAGIselRetries; while (1) { if (MatchScopes.empty()) { CannotYetSelect(NodeToMatch); - return 0; + return nullptr; } // Restore the interpreter state back to the point where the scope was @@ -2764,7 +3319,7 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, MatchedMemRefs.resize(LastScope.NumMatchedMemRefs); MatcherIndex = LastScope.FailIndex; - DEBUG(errs() << " Continuing at " << MatcherIndex << "\n"); + DEBUG(dbgs() << " Continuing at " << MatcherIndex << "\n"); InputChain = LastScope.InputChain; InputGlue = LastScope.InputGlue; @@ -2805,6 +3360,7 @@ void SelectionDAGISel::CannotYetSelect(SDNode *N) { N->getOpcode() != ISD::INTRINSIC_WO_CHAIN && N->getOpcode() != ISD::INTRINSIC_VOID) { N->printrFull(Msg, CurDAG); + Msg << "\nIn function: " << MF->getName(); } else { bool HasInputChain = N->getOperand(0).getValueType() == MVT::Other; unsigned iid =