X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FCodeGen%2FSelectionDAG%2FSelectionDAGISel.cpp;h=dc8044b7a7f6deb447107ed433448e958a320135;hb=d6379a993c7e40521bd5c8c6469e32697b4c41d1;hp=95ad4a6dc874ebaac22bef166e92e2bdc9c96b1d;hpb=c9af33c6854afe7b082af2d892ec5f05dfa383c7;p=oota-llvm.git diff --git a/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp b/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp index 95ad4a6dc87..dc8044b7a7f 100644 --- a/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp +++ b/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp @@ -14,9 +14,10 @@ #define DEBUG_TYPE "isel" #include "ScheduleDAGSDNodes.h" #include "SelectionDAGBuilder.h" -#include "FunctionLoweringInfo.h" +#include "llvm/CodeGen/FunctionLoweringInfo.h" #include "llvm/CodeGen/SelectionDAGISel.h" #include "llvm/Analysis/AliasAnalysis.h" +#include "llvm/Analysis/BranchProbabilityInfo.h" #include "llvm/Analysis/DebugInfo.h" #include "llvm/Constants.h" #include "llvm/Function.h" @@ -43,16 +44,21 @@ #include "llvm/Target/TargetLowering.h" #include "llvm/Target/TargetMachine.h" #include "llvm/Target/TargetOptions.h" +#include "llvm/Transforms/Utils/BasicBlockUtils.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 using namespace llvm; 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"); static cl::opt @@ -63,6 +69,11 @@ static cl::opt EnableFastISelAbort("fast-isel-abort", cl::Hidden, cl::desc("Enable abort calls when \"fast\" instruction fails")); +static cl::opt +UseMBPI("use-mbpi", + cl::desc("use Machine Branch Probability Info"), + cl::init(true), cl::Hidden); + #ifndef NDEBUG static cl::opt ViewDAGCombine1("view-dag-combine1-dags", cl::Hidden, @@ -132,14 +143,16 @@ namespace llvm { const TargetLowering &TLI = IS->getTargetLowering(); if (OptLevel == CodeGenOpt::None) - return createFastDAGScheduler(IS, OptLevel); + return createSourceListDAGScheduler(IS, OptLevel); if (TLI.getSchedulingPreference() == Sched::Latency) return createTDListDAGScheduler(IS, OptLevel); if (TLI.getSchedulingPreference() == Sched::RegPressure) return createBURRListDAGScheduler(IS, OptLevel); - assert(TLI.getSchedulingPreference() == Sched::Hybrid && + if (TLI.getSchedulingPreference() == Sched::Hybrid) + return createHybridListDAGScheduler(IS, OptLevel); + assert(TLI.getSchedulingPreference() == Sched::ILP && "Unknown sched type!"); - return createHybridListDAGScheduler(IS, OptLevel); + return createILPListDAGScheduler(IS, OptLevel); } } @@ -168,15 +181,19 @@ TargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI, // SelectionDAGISel code //===----------------------------------------------------------------------===// -SelectionDAGISel::SelectionDAGISel(const TargetMachine &tm, CodeGenOpt::Level OL) : - MachineFunctionPass(&ID), TM(tm), TLI(*tm.getTargetLowering()), +SelectionDAGISel::SelectionDAGISel(const TargetMachine &tm, + CodeGenOpt::Level OL) : + MachineFunctionPass(ID), TM(tm), TLI(*tm.getTargetLowering()), FuncInfo(new FunctionLoweringInfo(TLI)), CurDAG(new SelectionDAG(tm)), SDB(new SelectionDAGBuilder(*CurDAG, *FuncInfo, OL)), GFI(), OptLevel(OL), - DAGSize(0) -{} + DAGSize(0) { + initializeGCModuleInfoPass(*PassRegistry::getPassRegistry()); + initializeAliasAnalysisAnalysisGroup(*PassRegistry::getPassRegistry()); + initializeBranchProbabilityInfoPass(*PassRegistry::getPassRegistry()); + } SelectionDAGISel::~SelectionDAGISel() { delete SDB; @@ -189,40 +206,47 @@ void SelectionDAGISel::getAnalysisUsage(AnalysisUsage &AU) const { AU.addPreserved(); AU.addRequired(); AU.addPreserved(); + if (UseMBPI && OptLevel != CodeGenOpt::None) + AU.addRequired(); MachineFunctionPass::getAnalysisUsage(AU); } -/// FunctionCallsSetJmp - Return true if the function has a call to setjmp or -/// other function that gcc recognizes as "returning twice". This is used to -/// limit code-gen optimizations on the machine function. +/// SplitCriticalSideEffectEdges - Look for critical edges with a PHI value that +/// may trap on it. In this case we have to split the edge so that the path +/// through the predecessor block that doesn't go to the phi block doesn't +/// execute the possibly trapping instruction. /// -/// FIXME: Remove after is fixed. -static bool FunctionCallsSetJmp(const Function *F) { - const Module *M = F->getParent(); - static const char *ReturnsTwiceFns[] = { - "setjmp", - "sigsetjmp", - "setjmp_syscall", - "savectx", - "qsetjmp", - "vfork", - "getcontext" - }; -#define NUM_RETURNS_TWICE_FNS sizeof(ReturnsTwiceFns) / sizeof(const char *) - - for (unsigned I = 0; I < NUM_RETURNS_TWICE_FNS; ++I) - if (const Function *Callee = M->getFunction(ReturnsTwiceFns[I])) { - if (!Callee->use_empty()) - for (Value::const_use_iterator - I = Callee->use_begin(), E = Callee->use_end(); - I != E; ++I) - if (const CallInst *CI = dyn_cast(I)) - if (CI->getParent()->getParent() == F) - return true; - } +/// This is required for correctness, so it must be done at -O0. +/// +static void SplitCriticalSideEffectEdges(Function &Fn, Pass *SDISel) { + // 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; + + ReprocessBlock: + // For each block with a PHI node, check to see if any of the input values + // are potentially trapping constant expressions. Constant expressions are + // the only potentially trapping value that can occur as the argument to a + // PHI. + 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; + + // 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 + // edges: check to see if the pred has multiple successors. + BasicBlock *Pred = PN->getIncomingBlock(i); + if (Pred->getTerminator()->getNumSuccessors() == 1) + continue; - return false; -#undef NUM_RETURNS_TWICE_FNS + // Okay, we have to split this edge. + SplitCriticalEdge(Pred->getTerminator(), + GetSuccessorNumber(Pred, BB), SDISel, true); + goto ReprocessBlock; + } + } } bool SelectionDAGISel::runOnMachineFunction(MachineFunction &mf) { @@ -243,8 +267,16 @@ bool SelectionDAGISel::runOnMachineFunction(MachineFunction &mf) { DEBUG(dbgs() << "\n\n\n=== " << Fn.getName() << "\n"); + SplitCriticalSideEffectEdges(const_cast(Fn), this); + CurDAG->init(*MF); FuncInfo->set(Fn, *MF); + + if (UseMBPI && OptLevel != CodeGenOpt::None) + FuncInfo->BPI = &getAnalysis(); + else + FuncInfo->BPI = 0; + SDB->init(GFI, *AA); SelectAllBasicBlocks(Fn); @@ -259,7 +291,7 @@ bool SelectionDAGISel::runOnMachineFunction(MachineFunction &mf) { if (!FuncInfo->ArgDbgValues.empty()) for (MachineRegisterInfo::livein_iterator LI = RegInfo->livein_begin(), E = RegInfo->livein_end(); LI != E; ++LI) - if (LI->second) + if (LI->second) LiveInMap.insert(std::make_pair(LI->first, LI->second)); // Insert DBG_VALUE instructions for function arguments to the entry block. @@ -280,14 +312,37 @@ bool SelectionDAGISel::runOnMachineFunction(MachineFunction &mf) { if (LDI != LiveInMap.end()) { MachineInstr *Def = RegInfo->getVRegDef(LDI->second); MachineBasicBlock::iterator InsertPos = Def; - const MDNode *Variable = + const MDNode *Variable = MI->getOperand(MI->getNumOperands()-1).getMetadata(); unsigned Offset = MI->getOperand(1).getImm(); // Def is never a terminator here, so it is ok to increment InsertPos. - BuildMI(*EntryMBB, ++InsertPos, MI->getDebugLoc(), + BuildMI(*EntryMBB, ++InsertPos, MI->getDebugLoc(), TII.get(TargetOpcode::DBG_VALUE)) .addReg(LDI->second, RegState::Debug) .addImm(Offset).addMetadata(Variable); + + // 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();) { + 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; + } + if (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); + } } } @@ -300,7 +355,9 @@ bool SelectionDAGISel::runOnMachineFunction(MachineFunction &mf) { for (MachineBasicBlock::const_iterator II = MBB->begin(), IE = MBB->end(); II != IE; ++II) { const TargetInstrDesc &TID = TM.getInstrInfo()->get(II->getOpcode()); - if (II->isInlineAsm() || (TID.isCall() && !TID.isReturn())) { + + if ((TID.isCall() && !TID.isReturn()) || + II->isStackAligningInlineAsm()) { MFI->setHasCalls(true); goto done; } @@ -310,7 +367,27 @@ bool SelectionDAGISel::runOnMachineFunction(MachineFunction &mf) { } // Determine if there is a call to setjmp in the machine function. - MF->setCallsSetJmp(FunctionCallsSetJmp(&Fn)); + MF->setCallsSetJmp(Fn.callsFunctionThatReturnsTwice()); + + // Replace forward-declared registers with the registers containing + // the desired value. + MachineRegisterInfo &MRI = MF->getRegInfo(); + for (DenseMap::iterator + I = FuncInfo->RegFixups.begin(), E = FuncInfo->RegFixups.end(); + I != E; ++I) { + unsigned From = I->first; + unsigned To = I->second; + // If To is also scheduled to be replaced, find what its ultimate + // replacement is. + for (;;) { + DenseMap::iterator J = + FuncInfo->RegFixups.find(To); + if (J == E) break; + To = J->second; + } + // Replace it. + MRI.replaceRegWith(From, To); + } // Release function-specific state. SDB and CurDAG are already cleared // at this point. @@ -319,11 +396,9 @@ bool SelectionDAGISel::runOnMachineFunction(MachineFunction &mf) { return true; } -MachineBasicBlock * -SelectionDAGISel::SelectBasicBlock(MachineBasicBlock *BB, - BasicBlock::const_iterator Begin, - BasicBlock::const_iterator End, - bool &HadTailCall) { +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. @@ -336,39 +411,7 @@ SelectionDAGISel::SelectBasicBlock(MachineBasicBlock *BB, SDB->clear(); // Final step, emit the lowered DAG as machine code. - return CodeGenAndEmitDAG(BB); -} - -namespace { -/// WorkListRemover - This class is a DAGUpdateListener that removes any deleted -/// nodes from the worklist. -class SDOPsWorkListRemover : public SelectionDAG::DAGUpdateListener { - SmallVector &Worklist; - SmallPtrSet &InWorklist; -public: - SDOPsWorkListRemover(SmallVector &wl, - SmallPtrSet &inwl) - : Worklist(wl), InWorklist(inwl) {} - - void RemoveFromWorklist(SDNode *N) { - if (!InWorklist.erase(N)) return; - - SmallVector::iterator I = - std::find(Worklist.begin(), Worklist.end(), N); - assert(I != Worklist.end() && "Not in worklist"); - - *I = Worklist.back(); - Worklist.pop_back(); - } - - virtual void NodeDeleted(SDNode *N, SDNode *E) { - RemoveFromWorklist(N); - } - - virtual void NodeUpdated(SDNode *N) { - // Ignore updates. - } -}; + CodeGenAndEmitDAG(); } void SelectionDAGISel::ComputeLiveOutVRegInfo() { @@ -410,33 +453,28 @@ void SelectionDAGISel::ComputeLiveOutVRegInfo() { unsigned NumSignBits = CurDAG->ComputeNumSignBits(Src); Mask = APInt::getAllOnesValue(SrcVT.getSizeInBits()); CurDAG->ComputeMaskedBits(Src, Mask, KnownZero, KnownOne); - - // Only install this information if it tells us something. - if (NumSignBits != 1 || KnownZero != 0 || KnownOne != 0) { - DestReg -= TargetRegisterInfo::FirstVirtualRegister; - if (DestReg >= FuncInfo->LiveOutRegInfo.size()) - FuncInfo->LiveOutRegInfo.resize(DestReg+1); - FunctionLoweringInfo::LiveOutInfo &LOI = - FuncInfo->LiveOutRegInfo[DestReg]; - LOI.NumSignBits = NumSignBits; - LOI.KnownOne = KnownOne; - LOI.KnownZero = KnownZero; - } + FuncInfo->AddLiveOutRegInfo(DestReg, NumSignBits, KnownZero, KnownOne); } while (!Worklist.empty()); } -MachineBasicBlock *SelectionDAGISel::CodeGenAndEmitDAG(MachineBasicBlock *BB) { +void SelectionDAGISel::CodeGenAndEmitDAG() { std::string GroupName; if (TimePassesIsEnabled) GroupName = "Instruction Selection and Scheduling"; std::string BlockName; + int BlockNumber = -1; +#ifdef NDEBUG if (ViewDAGCombine1 || ViewLegalizeTypesDAGs || ViewLegalizeDAGs || ViewDAGCombine2 || ViewDAGCombineLT || ViewISelDAGs || ViewSchedDAGs || ViewSUnitDAGs) +#endif + { + BlockNumber = FuncInfo->MBB->getNumber(); BlockName = MF->getFunction()->getNameStr() + ":" + - BB->getBasicBlock()->getNameStr(); - - DEBUG(dbgs() << "Initial selection DAG:\n"; CurDAG->dump()); + FuncInfo->MBB->getBasicBlock()->getNameStr(); + } + DEBUG(dbgs() << "Initial selection DAG: BB#" << BlockNumber + << " '" << BlockName << "'\n"; CurDAG->dump()); if (ViewDAGCombine1) CurDAG->viewGraph("dag-combine1 input for " + BlockName); @@ -446,7 +484,8 @@ MachineBasicBlock *SelectionDAGISel::CodeGenAndEmitDAG(MachineBasicBlock *BB) { CurDAG->Combine(Unrestricted, *AA, OptLevel); } - DEBUG(dbgs() << "Optimized lowered selection DAG:\n"; CurDAG->dump()); + DEBUG(dbgs() << "Optimized lowered selection DAG: BB#" << BlockNumber + << " '" << BlockName << "'\n"; CurDAG->dump()); // Second step, hack on the DAG until it only uses operations and types that // the target supports. @@ -459,7 +498,8 @@ MachineBasicBlock *SelectionDAGISel::CodeGenAndEmitDAG(MachineBasicBlock *BB) { Changed = CurDAG->LegalizeTypes(); } - DEBUG(dbgs() << "Type-legalized selection DAG:\n"; CurDAG->dump()); + DEBUG(dbgs() << "Type-legalized selection DAG: BB#" << BlockNumber + << " '" << BlockName << "'\n"; CurDAG->dump()); if (Changed) { if (ViewDAGCombineLT) @@ -472,8 +512,8 @@ MachineBasicBlock *SelectionDAGISel::CodeGenAndEmitDAG(MachineBasicBlock *BB) { CurDAG->Combine(NoIllegalTypes, *AA, OptLevel); } - DEBUG(dbgs() << "Optimized type-legalized selection DAG:\n"; - CurDAG->dump()); + DEBUG(dbgs() << "Optimized type-legalized selection DAG: BB#" << BlockNumber + << " '" << BlockName << "'\n"; CurDAG->dump()); } { @@ -497,18 +537,19 @@ MachineBasicBlock *SelectionDAGISel::CodeGenAndEmitDAG(MachineBasicBlock *BB) { CurDAG->Combine(NoIllegalOperations, *AA, OptLevel); } - DEBUG(dbgs() << "Optimized vector-legalized selection DAG:\n"; - CurDAG->dump()); + DEBUG(dbgs() << "Optimized vector-legalized selection DAG: BB#" + << BlockNumber << " '" << BlockName << "'\n"; CurDAG->dump()); } if (ViewLegalizeDAGs) CurDAG->viewGraph("legalize input for " + BlockName); { NamedRegionTimer T("DAG Legalization", GroupName, TimePassesIsEnabled); - CurDAG->Legalize(OptLevel); + CurDAG->Legalize(); } - DEBUG(dbgs() << "Legalized selection DAG:\n"; CurDAG->dump()); + DEBUG(dbgs() << "Legalized selection DAG: BB#" << BlockNumber + << " '" << BlockName << "'\n"; CurDAG->dump()); if (ViewDAGCombine2) CurDAG->viewGraph("dag-combine2 input for " + BlockName); @@ -518,7 +559,8 @@ MachineBasicBlock *SelectionDAGISel::CodeGenAndEmitDAG(MachineBasicBlock *BB) { CurDAG->Combine(NoIllegalOperations, *AA, OptLevel); } - DEBUG(dbgs() << "Optimized legalized selection DAG:\n"; CurDAG->dump()); + DEBUG(dbgs() << "Optimized legalized selection DAG: BB#" << BlockNumber + << " '" << BlockName << "'\n"; CurDAG->dump()); if (OptLevel != CodeGenOpt::None) ComputeLiveOutVRegInfo(); @@ -532,7 +574,8 @@ MachineBasicBlock *SelectionDAGISel::CodeGenAndEmitDAG(MachineBasicBlock *BB) { DoInstructionSelection(); } - DEBUG(dbgs() << "Selected selection DAG:\n"; CurDAG->dump()); + DEBUG(dbgs() << "Selected selection DAG: BB#" << BlockNumber + << " '" << BlockName << "'\n"; CurDAG->dump()); if (ViewSchedDAGs) CurDAG->viewGraph("scheduler input for " + BlockName); @@ -541,18 +584,26 @@ MachineBasicBlock *SelectionDAGISel::CodeGenAndEmitDAG(MachineBasicBlock *BB) { { NamedRegionTimer T("Instruction Scheduling", GroupName, TimePassesIsEnabled); - Scheduler->Run(CurDAG, BB, BB->end()); + Scheduler->Run(CurDAG, FuncInfo->MBB, FuncInfo->InsertPt); } if (ViewSUnitDAGs) Scheduler->viewGraph(); // Emit machine code to BB. This can change 'BB' to the last block being // inserted into. + MachineBasicBlock *FirstMBB = FuncInfo->MBB, *LastMBB; { NamedRegionTimer T("Instruction Creation", GroupName, TimePassesIsEnabled); - BB = Scheduler->EmitSchedule(); + + LastMBB = FuncInfo->MBB = Scheduler->EmitSchedule(); + FuncInfo->InsertPt = Scheduler->InsertPos; } + // If the block was split, make sure we update any references that are used to + // update PHI nodes later on. + if (FirstMBB != LastMBB) + SDB->UpdateSplitBlock(FirstMBB, LastMBB); + // Free the scheduler state. { NamedRegionTimer T("Instruction Scheduling Cleanup", GroupName, @@ -562,27 +613,27 @@ MachineBasicBlock *SelectionDAGISel::CodeGenAndEmitDAG(MachineBasicBlock *BB) { // Free the SelectionDAG state, now that we're finished with it. CurDAG->clear(); - - return BB; } void SelectionDAGISel::DoInstructionSelection() { - DEBUG(errs() << "===== Instruction selection begins:\n"); + DEBUG(errs() << "===== Instruction selection begins: BB#" + << FuncInfo->MBB->getNumber() + << " '" << FuncInfo->MBB->getName() << "'\n"); PreprocessISelDAG(); - + // Select target instructions for the DAG. { // Number all nodes with a topological order and set DAGSize. DAGSize = CurDAG->AssignTopologicalOrder(); - + // Create a dummy node (which is not added to allnodes), that adds // 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()); ++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 @@ -594,19 +645,19 @@ void SelectionDAGISel::DoInstructionSelection() { // makes it theoretically possible to disable the DAGCombiner. if (Node->use_empty()) continue; - + SDNode *ResNode = Select(Node); - + // FIXME: This is pretty gross. 'Select' should be changed to not return // anything at all and this code should be nuked with a tactical strike. - + // If node should not be replaced, continue with the next one. if (ResNode == Node || Node->getOpcode() == ISD::DELETED_NODE) continue; // Replace node. 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. @@ -614,9 +665,9 @@ void SelectionDAGISel::DoInstructionSelection() { CurDAG->RemoveDeadNode(Node, &ISU); } } - + CurDAG->setRoot(Dummy.getValue()); - } + } DEBUG(errs() << "===== Instruction selection ends:\n"); @@ -625,21 +676,22 @@ void SelectionDAGISel::DoInstructionSelection() { /// PrepareEHLandingPad - Emit an EH_LABEL, set up live-in registers, and /// do other setup for EH landing-pad blocks. -void SelectionDAGISel::PrepareEHLandingPad(MachineBasicBlock *BB) { +void SelectionDAGISel::PrepareEHLandingPad() { // 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(BB); + MCSymbol *Label = MF->getMMI().addLandingPad(FuncInfo->MBB); const TargetInstrDesc &II = TM.getInstrInfo()->get(TargetOpcode::EH_LABEL); - BuildMI(BB, SDB->getCurDebugLoc(), II).addSym(Label); + BuildMI(*FuncInfo->MBB, FuncInfo->InsertPt, SDB->getCurDebugLoc(), II) + .addSym(Label); // Mark exception register as live in. unsigned Reg = TLI.getExceptionAddressRegister(); - if (Reg) BB->addLiveIn(Reg); + if (Reg) FuncInfo->MBB->addLiveIn(Reg); // Mark exception selector register as live in. Reg = TLI.getExceptionSelectorRegister(); - if (Reg) BB->addLiveIn(Reg); + if (Reg) FuncInfo->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 @@ -652,7 +704,7 @@ void SelectionDAGISel::PrepareEHLandingPad(MachineBasicBlock *BB) { // 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. - const BasicBlock *LLVMBB = BB->getBasicBlock(); + const BasicBlock *LLVMBB = FuncInfo->MBB->getBasicBlock(); const BranchInst *Br = dyn_cast(LLVMBB->getTerminator()); if (Br && Br->isUnconditional()) { // Critical edge? @@ -667,95 +719,234 @@ void SelectionDAGISel::PrepareEHLandingPad(MachineBasicBlock *BB) { } } + + +/// 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(); + } + + // 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; + + // 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; + + // 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()) + 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; + + // 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(); + + // Ask the target to try folding the load. + return FastIS->TryToFoldLoad(User, RI.getOperandNo(), LI); +} + +/// isFoldedOrDeadInstruction - Return true if the specified instruction is +/// side-effect free and is either dead or folded into a generated instruction. +/// Return false if it needs to be emitted. +static bool isFoldedOrDeadInstruction(const Instruction *I, + FunctionLoweringInfo *FuncInfo) { + return !I->mayWriteToMemory() && // Side-effecting instructions aren't folded. + !isa(I) && // Terminators aren't folded. + !isa(I) && // Debug instructions aren't folded. + !FuncInfo->isExportedInst(I); // Exported instrs must be computed. +} + void SelectionDAGISel::SelectAllBasicBlocks(const Function &Fn) { // Initialize the Fast-ISel state, if needed. FastISel *FastIS = 0; if (EnableFastISel) - FastIS = TLI.createFastISel(*MF, FuncInfo->ValueMap, FuncInfo->MBBMap, - FuncInfo->StaticAllocaMap, - FuncInfo->PHINodesToUpdate -#ifndef NDEBUG - , FuncInfo->CatchInfoLost -#endif - ); + FastIS = TLI.createFastISel(*FuncInfo); // Iterate over all basic blocks in the function. - for (Function::const_iterator I = Fn.begin(), E = Fn.end(); I != E; ++I) { - const BasicBlock *LLVMBB = &*I; - MachineBasicBlock *BB = FuncInfo->MBBMap[LLVMBB]; + ReversePostOrderTraversal RPOT(&Fn); + for (ReversePostOrderTraversal::rpo_iterator + I = RPOT.begin(), E = RPOT.end(); I != E; ++I) { + const BasicBlock *LLVMBB = *I; + + if (OptLevel != CodeGenOpt::None) { + bool AllPredsVisited = true; + for (const_pred_iterator PI = pred_begin(LLVMBB), PE = pred_end(LLVMBB); + PI != PE; ++PI) { + if (!FuncInfo->VisitedBBs.count(*PI)) { + AllPredsVisited = false; + break; + } + } + + if (AllPredsVisited) { + for (BasicBlock::const_iterator I = LLVMBB->begin(); + isa(I); ++I) + FuncInfo->ComputePHILiveOutRegInfo(cast(I)); + } else { + for (BasicBlock::const_iterator I = LLVMBB->begin(); + isa(I); ++I) + FuncInfo->InvalidatePHILiveOutRegInfo(cast(I)); + } + + 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 = Begin; + BasicBlock::const_iterator BI = End; + + 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); - // Setup an EH landing-pad block. - if (BB->isLandingPad()) - PrepareEHLandingPad(BB); - // Before doing SelectionDAG ISel, see if FastISel has been requested. if (FastIS) { + FastIS->startNewBlock(); + // 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(); - BB = CodeGenAndEmitDAG(BB); + 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)); + else + FastIS->setLastLocalValue(0); } - FastIS->startNewBlock(BB); + // Do FastISel on as many instructions as possible. - for (; BI != End; ++BI) { -#if 0 - // Defer instructions with no side effects; they'll be emitted - // on-demand later. - if (BI->isSafeToSpeculativelyExecute() && - !FuncInfo->isExportedInst(BI)) + for (; BI != Begin; --BI) { + const Instruction *Inst = llvm::prior(BI); + + // If we no longer require this instruction, skip it. + if (isFoldedOrDeadInstruction(Inst, FuncInfo)) continue; -#endif + + // 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(BI)) + if (FastIS->SelectInstruction(Inst)) { + ++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)); + if (!isFoldedOrDeadInstruction(BeforeInst, FuncInfo)) + break; + } + if (BeforeInst != Inst && isa(BeforeInst) && + BeforeInst->hasOneUse() && + TryToFoldFastISelLoad(cast(BeforeInst), Inst, FastIS)) + // If we succeeded, don't re-select the load. + BI = llvm::next(BasicBlock::const_iterator(BeforeInst)); continue; + } // Then handle certain instructions as single-LLVM-Instruction blocks. - if (isa(BI)) { + if (isa(Inst)) { ++NumFastIselFailures; if (EnableFastISelVerbose || EnableFastISelAbort) { dbgs() << "FastISel missed call: "; - BI->dump(); + Inst->dump(); } - if (!BI->getType()->isVoidTy() && !BI->use_empty()) { - unsigned &R = FuncInfo->ValueMap[BI]; + if (!Inst->getType()->isVoidTy() && !Inst->use_empty()) { + unsigned &R = FuncInfo->ValueMap[Inst]; if (!R) - R = FuncInfo->CreateRegs(BI->getType()); + R = FuncInfo->CreateRegs(Inst->getType()); } bool HadTailCall = false; - BB = SelectBasicBlock(BB, BI, llvm::next(BI), HadTailCall); + SelectBasicBlock(Inst, BI, HadTailCall); // If the call was emitted as a tail call, we're done with the block. if (HadTailCall) { - BI = End; + --BI; break; } - // If the instruction was codegen'd with multiple blocks, - // inform the FastISel object where to resume inserting. - FastIS->setCurrentBlock(BB); continue; } - // Otherwise, give up on FastISel for the rest of the block. - // For now, be a little lenient about non-branch terminators. - if (!isa(BI) || isa(BI)) { + if (isa(Inst) && !isa(Inst)) { + // Don't abort, and use a different message for terminator misses. + ++NumFastIselFailures; + if (EnableFastISelVerbose || EnableFastISelAbort) { + dbgs() << "FastISel missed terminator: "; + Inst->dump(); + } + } else { ++NumFastIselFailures; if (EnableFastISelVerbose || EnableFastISelAbort) { dbgs() << "FastISel miss: "; - BI->dump(); + Inst->dump(); } if (EnableFastISelAbort) // The "fast" selector couldn't handle something and bailed. @@ -764,25 +955,33 @@ void SelectionDAGISel::SelectAllBasicBlocks(const Function &Fn) { } break; } + + FastIS->recomputeInsertPt(); } - // Run SelectionDAG instruction selection on the remainder of the block - // not handled by FastISel. If FastISel is not run, this is the entire - // block. - if (BI != End) { + if (Begin != BI) + ++NumDAGBlocks; + else + ++NumFastIselBlocks; + + if (Begin != BI) { + // Run SelectionDAG instruction selection on the remainder of the block + // not handled by FastISel. If FastISel is not run, this is the entire + // block. bool HadTailCall; - BB = SelectBasicBlock(BB, BI, End, HadTailCall); + SelectBasicBlock(Begin, BI, HadTailCall); } - FinishBasicBlock(BB); + FinishBasicBlock(); FuncInfo->PHINodesToUpdate.clear(); } delete FastIS; + SDB->clearDanglingDebugInfo(); } void -SelectionDAGISel::FinishBasicBlock(MachineBasicBlock *BB) { +SelectionDAGISel::FinishBasicBlock() { DEBUG(dbgs() << "Total amount of phi nodes to update: " << FuncInfo->PHINodesToUpdate.size() << "\n"; @@ -800,11 +999,11 @@ SelectionDAGISel::FinishBasicBlock(MachineBasicBlock *BB) { MachineInstr *PHI = FuncInfo->PHINodesToUpdate[i].first; assert(PHI->isPHI() && "This is not a machine PHI node that we are updating!"); - if (!BB->isSuccessor(PHI->getParent())) + if (!FuncInfo->MBB->isSuccessor(PHI->getParent())) continue; PHI->addOperand( MachineOperand::CreateReg(FuncInfo->PHINodesToUpdate[i].second, false)); - PHI->addOperand(MachineOperand::CreateMBB(BB)); + PHI->addOperand(MachineOperand::CreateMBB(FuncInfo->MBB)); } return; } @@ -813,33 +1012,37 @@ SelectionDAGISel::FinishBasicBlock(MachineBasicBlock *BB) { // Lower header first, if it wasn't already lowered if (!SDB->BitTestCases[i].Emitted) { // Set the current basic block to the mbb we wish to insert the code into - BB = SDB->BitTestCases[i].Parent; + FuncInfo->MBB = SDB->BitTestCases[i].Parent; + FuncInfo->InsertPt = FuncInfo->MBB->end(); // Emit the code - SDB->visitBitTestHeader(SDB->BitTestCases[i], BB); + SDB->visitBitTestHeader(SDB->BitTestCases[i], FuncInfo->MBB); CurDAG->setRoot(SDB->getRoot()); SDB->clear(); - BB = CodeGenAndEmitDAG(BB); + CodeGenAndEmitDAG(); } for (unsigned j = 0, ej = SDB->BitTestCases[i].Cases.size(); j != ej; ++j) { // Set the current basic block to the mbb we wish to insert the code into - BB = SDB->BitTestCases[i].Cases[j].ThisBB; + FuncInfo->MBB = SDB->BitTestCases[i].Cases[j].ThisBB; + FuncInfo->InsertPt = FuncInfo->MBB->end(); // Emit the code if (j+1 != ej) - SDB->visitBitTestCase(SDB->BitTestCases[i].Cases[j+1].ThisBB, + SDB->visitBitTestCase(SDB->BitTestCases[i], + SDB->BitTestCases[i].Cases[j+1].ThisBB, SDB->BitTestCases[i].Reg, SDB->BitTestCases[i].Cases[j], - BB); + FuncInfo->MBB); else - SDB->visitBitTestCase(SDB->BitTestCases[i].Default, + SDB->visitBitTestCase(SDB->BitTestCases[i], + SDB->BitTestCases[i].Default, SDB->BitTestCases[i].Reg, SDB->BitTestCases[i].Cases[j], - BB); + FuncInfo->MBB); CurDAG->setRoot(SDB->getRoot()); SDB->clear(); - BB = CodeGenAndEmitDAG(BB); + CodeGenAndEmitDAG(); } // Update PHI Nodes @@ -884,22 +1087,24 @@ SelectionDAGISel::FinishBasicBlock(MachineBasicBlock *BB) { // Lower header first, if it wasn't already lowered if (!SDB->JTCases[i].first.Emitted) { // Set the current basic block to the mbb we wish to insert the code into - BB = SDB->JTCases[i].first.HeaderBB; + FuncInfo->MBB = SDB->JTCases[i].first.HeaderBB; + FuncInfo->InsertPt = FuncInfo->MBB->end(); // Emit the code SDB->visitJumpTableHeader(SDB->JTCases[i].second, SDB->JTCases[i].first, - BB); + FuncInfo->MBB); CurDAG->setRoot(SDB->getRoot()); SDB->clear(); - BB = CodeGenAndEmitDAG(BB); + CodeGenAndEmitDAG(); } // Set the current basic block to the mbb we wish to insert the code into - BB = SDB->JTCases[i].second.MBB; + FuncInfo->MBB = SDB->JTCases[i].second.MBB; + FuncInfo->InsertPt = FuncInfo->MBB->end(); // Emit the code SDB->visitJumpTable(SDB->JTCases[i].second); CurDAG->setRoot(SDB->getRoot()); SDB->clear(); - BB = CodeGenAndEmitDAG(BB); + CodeGenAndEmitDAG(); // Update PHI Nodes for (unsigned pi = 0, pe = FuncInfo->PHINodesToUpdate.size(); @@ -917,11 +1122,11 @@ SelectionDAGISel::FinishBasicBlock(MachineBasicBlock *BB) { (MachineOperand::CreateMBB(SDB->JTCases[i].first.HeaderBB)); } // JT BB. Just iterate over successors here - if (BB->isSuccessor(PHIBB)) { + if (FuncInfo->MBB->isSuccessor(PHIBB)) { PHI->addOperand (MachineOperand::CreateReg(FuncInfo->PHINodesToUpdate[pi].second, false)); - PHI->addOperand(MachineOperand::CreateMBB(BB)); + PHI->addOperand(MachineOperand::CreateMBB(FuncInfo->MBB)); } } } @@ -933,10 +1138,10 @@ SelectionDAGISel::FinishBasicBlock(MachineBasicBlock *BB) { MachineInstr *PHI = FuncInfo->PHINodesToUpdate[i].first; assert(PHI->isPHI() && "This is not a machine PHI node that we are updating!"); - if (BB->isSuccessor(PHI->getParent())) { + if (FuncInfo->MBB->isSuccessor(PHI->getParent())) { PHI->addOperand( MachineOperand::CreateReg(FuncInfo->PHINodesToUpdate[i].second, false)); - PHI->addOperand(MachineOperand::CreateMBB(BB)); + PHI->addOperand(MachineOperand::CreateMBB(FuncInfo->MBB)); } } @@ -944,7 +1149,8 @@ SelectionDAGISel::FinishBasicBlock(MachineBasicBlock *BB) { // additional DAGs necessary. for (unsigned i = 0, e = SDB->SwitchCases.size(); i != e; ++i) { // Set the current basic block to the mbb we wish to insert the code into - MachineBasicBlock *ThisBB = BB = SDB->SwitchCases[i].ThisBB; + FuncInfo->MBB = SDB->SwitchCases[i].ThisBB; + FuncInfo->InsertPt = FuncInfo->MBB->end(); // Determine the unique successors. SmallVector Succs; @@ -952,23 +1158,28 @@ SelectionDAGISel::FinishBasicBlock(MachineBasicBlock *BB) { if (SDB->SwitchCases[i].TrueBB != SDB->SwitchCases[i].FalseBB) Succs.push_back(SDB->SwitchCases[i].FalseBB); - // Emit the code. Note that this could result in ThisBB being split, so - // we need to check for updates. - SDB->visitSwitchCase(SDB->SwitchCases[i], BB); + // Emit the code. Note that this could result in FuncInfo->MBB being split. + SDB->visitSwitchCase(SDB->SwitchCases[i], FuncInfo->MBB); CurDAG->setRoot(SDB->getRoot()); SDB->clear(); - ThisBB = CodeGenAndEmitDAG(BB); + CodeGenAndEmitDAG(); + + // Remember the last block, now that any splitting is done, for use in + // populating PHI nodes in successors. + MachineBasicBlock *ThisBB = FuncInfo->MBB; // Handle any PHI nodes in successors of this chunk, as if we were coming // from the original BB before switch expansion. Note that PHI nodes can // occur multiple times in PHINodesToUpdate. We have to be very careful to // handle them the right number of times. for (unsigned i = 0, e = Succs.size(); i != e; ++i) { - BB = Succs[i]; - // BB may have been removed from the CFG if a branch was constant folded. - if (ThisBB->isSuccessor(BB)) { - for (MachineBasicBlock::iterator Phi = BB->begin(); - Phi != BB->end() && Phi->isPHI(); + FuncInfo->MBB = Succs[i]; + FuncInfo->InsertPt = FuncInfo->MBB->end(); + // 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) { // This value for this PHI node is recorded in PHINodesToUpdate. for (unsigned pn = 0; ; ++pn) { @@ -1005,10 +1216,6 @@ ScheduleDAGSDNodes *SelectionDAGISel::CreateScheduler() { return Ctor(this, OptLevel); } -ScheduleHazardRecognizer *SelectionDAGISel::CreateTargetHazardRecognizer() { - return new ScheduleHazardRecognizer(); -} - //===----------------------------------------------------------------------===// // Helper functions used by the generated instruction selector. //===----------------------------------------------------------------------===// @@ -1088,11 +1295,11 @@ SelectInlineAsmMemoryOperands(std::vector &Ops) { Ops.push_back(InOps[InlineAsm::Op_InputChain]); // 0 Ops.push_back(InOps[InlineAsm::Op_AsmString]); // 1 Ops.push_back(InOps[InlineAsm::Op_MDNode]); // 2, !srcloc - Ops.push_back(InOps[InlineAsm::Op_IsAlignStack]); // 3 + Ops.push_back(InOps[InlineAsm::Op_ExtraInfo]); // 3 (SideEffect, AlignStack) unsigned i = InlineAsm::Op_FirstOperand, e = InOps.size(); - if (InOps[e-1].getValueType() == MVT::Flag) - --e; // Don't process a flag operand if it is here. + if (InOps[e-1].getValueType() == MVT::Glue) + --e; // Don't process a glue operand if it is here. while (i != e) { unsigned Flags = cast(InOps[i])->getZExtValue(); @@ -1119,15 +1326,15 @@ SelectInlineAsmMemoryOperands(std::vector &Ops) { } } - // Add the flag input back if present. + // Add the glue input back if present. if (e != InOps.size()) Ops.push_back(InOps.back()); } -/// findFlagUse - Return use of EVT::Flag value produced by the specified +/// findGlueUse - Return use of MVT::Glue value produced by the specified /// SDNode. /// -static SDNode *findFlagUse(SDNode *N) { +static SDNode *findGlueUse(SDNode *N) { unsigned FlagResNo = N->getNumValues()-1; for (SDNode::use_iterator I = N->use_begin(), E = N->use_end(); I != E; ++I) { SDUse &Use = I.getUse(); @@ -1149,11 +1356,11 @@ static bool findNonImmUse(SDNode *Use, SDNode* Def, SDNode *ImmedUse, // never find it. // // The Use may be -1 (unassigned) if it is a newly allocated node. This can - // happen because we scan down to newly selected nodes in the case of flag + // happen because we scan down to newly selected nodes in the case of glue // uses. if ((Use->getNodeId() < Def->getNodeId() && Use->getNodeId() != -1)) return false; - + // 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)) @@ -1163,7 +1370,7 @@ static bool findNonImmUse(SDNode *Use, SDNode* Def, SDNode *ImmedUse, // Ignore chain uses, they are validated by HandleMergeInputChains. if (Use->getOperand(i).getValueType() == MVT::Other && IgnoreChains) continue; - + SDNode *N = Use->getOperand(i).getNode(); if (N == Def) { if (Use == ImmedUse || Use == Root) @@ -1210,8 +1417,8 @@ bool SelectionDAGISel::IsLegalToFold(SDValue N, SDNode *U, SDNode *Root, // // * indicates nodes to be folded together. // - // If Root produces a flag, then it gets (even more) interesting. Since it - // will be "glued" together with its flag use in the scheduler, we need to + // If Root produces glue, then it gets (even more) interesting. Since it + // will be "glued" together with its glue use in the scheduler, we need to // check if it might reach N. // // [N*] // @@ -1229,30 +1436,30 @@ bool SelectionDAGISel::IsLegalToFold(SDValue N, SDNode *U, SDNode *Root, // ^ / // // f / // // | / // - // [FU] // + // [GU] // // - // If FU (flag use) indirectly reaches N (the load), and Root folds N - // (call it Fold), then X is a predecessor of FU and a successor of - // Fold. But since Fold and FU are flagged together, this will create + // If GU (glue use) indirectly reaches N (the load), and Root folds N + // (call it Fold), then X is a predecessor of GU and a successor of + // Fold. But since Fold and GU are glued together, this will create // a cycle in the scheduling graph. - // If the node has flags, walk down the graph to the "lowest" node in the - // flagged set. + // If the node has glue, walk down the graph to the "lowest" node in the + // glueged set. EVT VT = Root->getValueType(Root->getNumValues()-1); - while (VT == MVT::Flag) { - SDNode *FU = findFlagUse(Root); - if (FU == NULL) + while (VT == MVT::Glue) { + SDNode *GU = findGlueUse(Root); + if (GU == NULL) break; - Root = FU; + Root = GU; VT = Root->getValueType(Root->getNumValues()-1); - - // If our query node has a flag result with a use, we've walked up it. If + + // If our query node has a glue result with a use, we've walked up it. If // the user (which has already been selected) has a chain or indirectly uses // the chain, our WalkChainUsers predicate will not consider it. Because of // this, we cannot ignore chains in this predicate. IgnoreChains = false; } - + SmallPtrSet Visited; return !findNonImmUse(Root, N.getNode(), U, Root, Visited, IgnoreChains); @@ -1261,10 +1468,10 @@ bool SelectionDAGISel::IsLegalToFold(SDValue N, SDNode *U, SDNode *Root, SDNode *SelectionDAGISel::Select_INLINEASM(SDNode *N) { std::vector Ops(N->op_begin(), N->op_end()); SelectInlineAsmMemoryOperands(Ops); - + std::vector VTs; VTs.push_back(MVT::Other); - VTs.push_back(MVT::Flag); + VTs.push_back(MVT::Glue); SDValue New = CurDAG->getNode(ISD::INLINEASM, N->getDebugLoc(), VTs, &Ops[0], Ops.size()); New->setNodeId(-1); @@ -1276,11 +1483,11 @@ SDNode *SelectionDAGISel::Select_UNDEF(SDNode *N) { } /// GetVBR - decode a vbr encoding whose top bit is set. -ALWAYS_INLINE static uint64_t +LLVM_ATTRIBUTE_ALWAYS_INLINE static uint64_t GetVBR(uint64_t Val, const unsigned char *MatcherTable, unsigned &Idx) { assert(Val >= 128 && "Not a VBR"); Val &= 127; // Remove first vbr bit. - + unsigned Shift = 7; uint64_t NextBits; do { @@ -1288,25 +1495,25 @@ GetVBR(uint64_t Val, const unsigned char *MatcherTable, unsigned &Idx) { Val |= (NextBits&127) << Shift; Shift += 7; } while (NextBits & 128); - + return Val; } -/// UpdateChainsAndFlags - When a match is complete, this method updates uses of -/// interior flag and chain results to use the new flag and chain results. +/// UpdateChainsAndGlue - When a match is complete, this method updates uses of +/// interior glue and chain results to use the new glue and chain results. void SelectionDAGISel:: -UpdateChainsAndFlags(SDNode *NodeToMatch, SDValue InputChain, - const SmallVectorImpl &ChainNodesMatched, - SDValue InputFlag, - const SmallVectorImpl &FlagResultNodesMatched, - bool isMorphNodeTo) { +UpdateChainsAndGlue(SDNode *NodeToMatch, SDValue InputChain, + const SmallVectorImpl &ChainNodesMatched, + SDValue InputGlue, + const SmallVectorImpl &GlueResultNodesMatched, + bool isMorphNodeTo) { SmallVector NowDeadNodes; - + ISelUpdater ISU(ISelPosition); // Now that all the normal results are replaced, we replace the chain and - // flag results if present. + // glue results if present. if (!ChainNodesMatched.empty()) { assert(InputChain.getNode() != 0 && "Matched input chains but didn't produce a chain"); @@ -1314,55 +1521,55 @@ UpdateChainsAndFlags(SDNode *NodeToMatch, SDValue InputChain, // Replace all the chain results with the final chain we ended up with. for (unsigned i = 0, e = ChainNodesMatched.size(); i != e; ++i) { SDNode *ChainNode = ChainNodesMatched[i]; - + // If this node was already deleted, don't look at it. if (ChainNode->getOpcode() == ISD::DELETED_NODE) continue; - + // Don't replace the results of the root node if we're doing a // MorphNodeTo. if (ChainNode == NodeToMatch && isMorphNodeTo) continue; - + SDValue ChainVal = SDValue(ChainNode, ChainNode->getNumValues()-1); - if (ChainVal.getValueType() == MVT::Flag) + if (ChainVal.getValueType() == MVT::Glue) ChainVal = ChainVal.getValue(ChainVal->getNumValues()-2); assert(ChainVal.getValueType() == MVT::Other && "Not a chain?"); CurDAG->ReplaceAllUsesOfValueWith(ChainVal, InputChain, &ISU); - + // If the node became dead and we haven't already seen it, delete it. if (ChainNode->use_empty() && !std::count(NowDeadNodes.begin(), NowDeadNodes.end(), ChainNode)) NowDeadNodes.push_back(ChainNode); } } - - // If the result produces a flag, update any flag results in the matched - // pattern with the flag result. - if (InputFlag.getNode() != 0) { + + // If the result produces glue, update any glue results in the matched + // pattern with the glue result. + if (InputGlue.getNode() != 0) { // Handle any interior nodes explicitly marked. - for (unsigned i = 0, e = FlagResultNodesMatched.size(); i != e; ++i) { - SDNode *FRN = FlagResultNodesMatched[i]; - + for (unsigned i = 0, e = GlueResultNodesMatched.size(); i != e; ++i) { + SDNode *FRN = GlueResultNodesMatched[i]; + // If this node was already deleted, don't look at it. if (FRN->getOpcode() == ISD::DELETED_NODE) continue; - - assert(FRN->getValueType(FRN->getNumValues()-1) == MVT::Flag && - "Doesn't have a flag result"); + + assert(FRN->getValueType(FRN->getNumValues()-1) == MVT::Glue && + "Doesn't have a glue result"); CurDAG->ReplaceAllUsesOfValueWith(SDValue(FRN, FRN->getNumValues()-1), - InputFlag, &ISU); - + InputGlue, &ISU); + // If the node became dead and we haven't already seen it, delete it. if (FRN->use_empty() && !std::count(NowDeadNodes.begin(), NowDeadNodes.end(), FRN)) NowDeadNodes.push_back(FRN); } } - + if (!NowDeadNodes.empty()) CurDAG->RemoveDeadNodes(NowDeadNodes, &ISU); - + DEBUG(errs() << "ISEL: Match complete!\n"); } @@ -1381,17 +1588,17 @@ 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 +static ChainResult WalkChainUsers(SDNode *ChainedNode, SmallVectorImpl &ChainedNodesInPattern, SmallVectorImpl &InteriorChainedNodes) { ChainResult Result = CR_Simple; - + for (SDNode::use_iterator UI = ChainedNode->use_begin(), E = ChainedNode->use_end(); UI != E; ++UI) { // Make sure the use is of the chain, not some other value we produce. if (UI.getUse().getValueType() != MVT::Other) continue; - + SDNode *User = *UI; // If we see an already-selected machine node, then we've gone beyond the @@ -1400,7 +1607,7 @@ WalkChainUsers(SDNode *ChainedNode, 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 || @@ -1426,7 +1633,7 @@ WalkChainUsers(SDNode *ChainedNode, if (!std::count(ChainedNodesInPattern.begin(), ChainedNodesInPattern.end(), User)) return CR_InducesCycle; - + // Otherwise we found a node that is part of our pattern. For example in: // x = load ptr // y = x+4 @@ -1438,7 +1645,7 @@ WalkChainUsers(SDNode *ChainedNode, InteriorChainedNodes.push_back(User); continue; } - + // If we found a TokenFactor, there are two cases to consider: first if the // TokenFactor is just hanging "below" the pattern we're matching (i.e. no // uses of the TF are in our pattern) we just want to ignore it. Second, @@ -1475,7 +1682,7 @@ WalkChainUsers(SDNode *ChainedNode, case CR_LeadsToInteriorNode: break; // Otherwise, keep processing. } - + // Okay, we know we're in the interesting interior case. The TokenFactor // is now going to be considered part of the pattern so that we rewrite its // uses (it may have uses that are not part of the pattern) with the @@ -1486,7 +1693,7 @@ WalkChainUsers(SDNode *ChainedNode, InteriorChainedNodes.push_back(User); continue; } - + return Result; } @@ -1508,7 +1715,7 @@ HandleMergeInputChains(SmallVectorImpl &ChainNodesMatched, InteriorChainedNodes) == CR_InducesCycle) return SDValue(); // Would induce a cycle. } - + // Okay, we have walked all the matched nodes and collected TokenFactor nodes // that we are interested in. Form our input TokenFactor node. SmallVector InputChains; @@ -1519,14 +1726,14 @@ HandleMergeInputChains(SmallVectorImpl &ChainNodesMatched, if (N->getOpcode() != ISD::TokenFactor) { if (std::count(InteriorChainedNodes.begin(),InteriorChainedNodes.end(),N)) continue; - + // Otherwise, add the input chain. SDValue InChain = ChainNodesMatched[i]->getOperand(0); assert(InChain.getValueType() == MVT::Other && "Not a chain"); InputChains.push_back(InChain); continue; } - + // If we have a token factor, we want to add all inputs of the token factor // that are not part of the pattern we're matching. for (unsigned op = 0, e = N->getNumOperands(); op != e; ++op) { @@ -1535,13 +1742,13 @@ HandleMergeInputChains(SmallVectorImpl &ChainNodesMatched, InputChains.push_back(N->getOperand(op)); } } - + SDValue Res; if (InputChains.size() == 1) return InputChains[0]; return CurDAG->getNode(ISD::TokenFactor, ChainNodesMatched[0]->getDebugLoc(), MVT::Other, &InputChains[0], InputChains.size()); -} +} /// MorphNode - Handle morphing a node in place for the selector. SDNode *SelectionDAGISel:: @@ -1549,15 +1756,15 @@ MorphNode(SDNode *Node, unsigned TargetOpc, SDVTList VTList, const SDValue *Ops, unsigned NumOps, 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 flags and chains as well. + // adding a chain) and the input could have glue and chains as well. // In this case we need to shift the operands down. // FIXME: This is a horrible hack and broken in obscure cases, no worse // than the old isel though. - int OldFlagResultNo = -1, OldChainResultNo = -1; + int OldGlueResultNo = -1, OldChainResultNo = -1; unsigned NTMNumResults = Node->getNumValues(); - if (Node->getValueType(NTMNumResults-1) == MVT::Flag) { - OldFlagResultNo = NTMNumResults-1; + if (Node->getValueType(NTMNumResults-1) == MVT::Glue) { + OldGlueResultNo = NTMNumResults-1; if (NTMNumResults != 1 && Node->getValueType(NTMNumResults-2) == MVT::Other) OldChainResultNo = NTMNumResults-2; @@ -1578,54 +1785,55 @@ MorphNode(SDNode *Node, unsigned TargetOpc, SDVTList VTList, } unsigned ResNumResults = Res->getNumValues(); - // Move the flag if needed. - if ((EmitNodeInfo & OPFL_FlagOutput) && OldFlagResultNo != -1 && - (unsigned)OldFlagResultNo != ResNumResults-1) - CurDAG->ReplaceAllUsesOfValueWith(SDValue(Node, OldFlagResultNo), + // Move the glue if needed. + if ((EmitNodeInfo & OPFL_GlueOutput) && OldGlueResultNo != -1 && + (unsigned)OldGlueResultNo != ResNumResults-1) + CurDAG->ReplaceAllUsesOfValueWith(SDValue(Node, OldGlueResultNo), SDValue(Res, ResNumResults-1)); - if ((EmitNodeInfo & OPFL_FlagOutput) != 0) + if ((EmitNodeInfo & OPFL_GlueOutput) != 0) --ResNumResults; // Move the chain reference if needed. if ((EmitNodeInfo & OPFL_Chain) && OldChainResultNo != -1 && (unsigned)OldChainResultNo != ResNumResults-1) - CurDAG->ReplaceAllUsesOfValueWith(SDValue(Node, OldChainResultNo), + CurDAG->ReplaceAllUsesOfValueWith(SDValue(Node, OldChainResultNo), SDValue(Res, ResNumResults-1)); // Otherwise, no replacement happened because the node already exists. Replace // Uses of the old node with the new one. if (Res != Node) CurDAG->ReplaceAllUsesWith(Node, Res); - + return Res; } /// CheckPatternPredicate - Implements OP_CheckPatternPredicate. -ALWAYS_INLINE static bool +LLVM_ATTRIBUTE_ALWAYS_INLINE static bool CheckSame(const unsigned char *MatcherTable, unsigned &MatcherIndex, - SDValue N, const SmallVectorImpl &RecordedNodes) { + SDValue N, + const SmallVectorImpl > &RecordedNodes) { // Accept if it is exactly the same as a previously recorded node. unsigned RecNo = MatcherTable[MatcherIndex++]; assert(RecNo < RecordedNodes.size() && "Invalid CheckSame"); - return N == RecordedNodes[RecNo]; + return N == RecordedNodes[RecNo].first; } - + /// CheckPatternPredicate - Implements OP_CheckPatternPredicate. -ALWAYS_INLINE static bool +LLVM_ATTRIBUTE_ALWAYS_INLINE static bool CheckPatternPredicate(const unsigned char *MatcherTable, unsigned &MatcherIndex, SelectionDAGISel &SDISel) { return SDISel.CheckPatternPredicate(MatcherTable[MatcherIndex++]); } /// CheckNodePredicate - Implements OP_CheckNodePredicate. -ALWAYS_INLINE static bool +LLVM_ATTRIBUTE_ALWAYS_INLINE static bool CheckNodePredicate(const unsigned char *MatcherTable, unsigned &MatcherIndex, SelectionDAGISel &SDISel, SDNode *N) { return SDISel.CheckNodePredicate(N, MatcherTable[MatcherIndex++]); } -ALWAYS_INLINE static bool +LLVM_ATTRIBUTE_ALWAYS_INLINE static bool CheckOpcode(const unsigned char *MatcherTable, unsigned &MatcherIndex, SDNode *N) { uint16_t Opc = MatcherTable[MatcherIndex++]; @@ -1633,17 +1841,17 @@ CheckOpcode(const unsigned char *MatcherTable, unsigned &MatcherIndex, return N->getOpcode() == Opc; } -ALWAYS_INLINE static bool +LLVM_ATTRIBUTE_ALWAYS_INLINE static bool CheckType(const unsigned char *MatcherTable, unsigned &MatcherIndex, 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(); } -ALWAYS_INLINE static bool +LLVM_ATTRIBUTE_ALWAYS_INLINE static bool CheckChildType(const unsigned char *MatcherTable, unsigned &MatcherIndex, SDValue N, const TargetLowering &TLI, unsigned ChildNo) { @@ -1653,57 +1861,57 @@ CheckChildType(const unsigned char *MatcherTable, unsigned &MatcherIndex, } -ALWAYS_INLINE static bool +LLVM_ATTRIBUTE_ALWAYS_INLINE static bool CheckCondCode(const unsigned char *MatcherTable, unsigned &MatcherIndex, SDValue N) { return cast(N)->get() == (ISD::CondCode)MatcherTable[MatcherIndex++]; } -ALWAYS_INLINE static bool +LLVM_ATTRIBUTE_ALWAYS_INLINE static bool CheckValueType(const unsigned char *MatcherTable, unsigned &MatcherIndex, 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(); } -ALWAYS_INLINE static bool +LLVM_ATTRIBUTE_ALWAYS_INLINE static bool CheckInteger(const unsigned char *MatcherTable, unsigned &MatcherIndex, SDValue N) { int64_t Val = MatcherTable[MatcherIndex++]; if (Val & 128) Val = GetVBR(Val, MatcherTable, MatcherIndex); - + ConstantSDNode *C = dyn_cast(N); return C != 0 && C->getSExtValue() == Val; } -ALWAYS_INLINE static bool +LLVM_ATTRIBUTE_ALWAYS_INLINE static bool CheckAndImm(const unsigned char *MatcherTable, unsigned &MatcherIndex, SDValue N, SelectionDAGISel &SDISel) { int64_t Val = MatcherTable[MatcherIndex++]; if (Val & 128) Val = GetVBR(Val, MatcherTable, 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); } -ALWAYS_INLINE static bool +LLVM_ATTRIBUTE_ALWAYS_INLINE static bool CheckOrImm(const unsigned char *MatcherTable, unsigned &MatcherIndex, SDValue N, SelectionDAGISel &SDISel) { int64_t Val = MatcherTable[MatcherIndex++]; if (Val & 128) Val = GetVBR(Val, MatcherTable, 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); } @@ -1713,11 +1921,11 @@ CheckOrImm(const unsigned char *MatcherTable, unsigned &MatcherIndex, /// fail, set Result=true and return anything. If the current predicate is /// known to pass, set Result=false and return the MatcherIndex to continue /// with. If the current predicate is unknown, set Result=false and return the -/// MatcherIndex to continue with. +/// MatcherIndex to continue with. static unsigned IsPredicateKnownToFail(const unsigned char *Table, unsigned Index, SDValue N, bool &Result, SelectionDAGISel &SDISel, - SmallVectorImpl &RecordedNodes){ + SmallVectorImpl > &RecordedNodes) { switch (Table[Index++]) { default: Result = false; @@ -1771,21 +1979,21 @@ namespace { struct MatchScope { /// FailIndex - If this match fails, this is the index to continue with. unsigned FailIndex; - + /// NodeStack - The node stack when the scope was formed. SmallVector NodeStack; - + /// NumRecordedNodes - The number of recorded nodes when the scope was formed. unsigned NumRecordedNodes; - + /// NumMatchedMemRefs - The number of matched memref entries. unsigned NumMatchedMemRefs; - - /// InputChain/InputFlag - The current chain/flag - SDValue InputChain, InputFlag; + + /// InputChain/InputGlue - The current chain/glue + SDValue InputChain, InputGlue; /// HasChainNodesMatched - True if the ChainNodesMatched list is non-empty. - bool HasChainNodesMatched, HasFlagResultNodesMatched; + bool HasChainNodesMatched, HasGlueResultNodesMatched; }; } @@ -1827,7 +2035,7 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, case ISD::INLINEASM: return Select_INLINEASM(NodeToMatch); case ISD::UNDEF: return Select_UNDEF(NodeToMatch); } - + assert(!NodeToMatch->isMachineOpcode() && "Node already selected!"); // Set up the node stack with NodeToMatch as the only node on the stack. @@ -1838,37 +2046,38 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, // MatchScopes - Scopes used when matching, if a match failure happens, this // indicates where to continue checking. SmallVector MatchScopes; - + // RecordedNodes - This is the set of nodes that have been recorded by the - // state machine. - SmallVector RecordedNodes; - + // state machine. The second value is the parent of the node, or null if the + // root is recorded. + SmallVector, 8> RecordedNodes; + // MatchedMemRefs - This is the set of MemRef's we've seen in the input // pattern. SmallVector MatchedMemRefs; - - // These are the current input chain and flag for use when generating nodes. + + // These are the current input chain and glue for use when generating nodes. // Various Emit operations change these. For example, emitting a copytoreg // uses and updates these. - SDValue InputChain, InputFlag; - + SDValue InputChain, InputGlue; + // ChainNodesMatched - If a pattern matches nodes that have input/output // chains, the OPC_EmitMergeInputChains operation is emitted which indicates // which ones they are. The result is captured into this list so that we can // update the chain results when the pattern is complete. SmallVector ChainNodesMatched; - SmallVector FlagResultNodesMatched; - + SmallVector GlueResultNodesMatched; + DEBUG(errs() << "ISEL: Starting pattern match on root node: "; NodeToMatch->dump(CurDAG); errs() << '\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 // accelerate the first lookup (which is guaranteed to be hot) with the // OpcodeOffset table. unsigned MatcherIndex = 0; - + if (!OpcodeOffset.empty()) { // Already computed the OpcodeOffset table, just index into it. if (N.getOpcode() < OpcodeOffset.size()) @@ -1900,7 +2109,7 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, if (N.getOpcode() < OpcodeOffset.size()) MatcherIndex = OpcodeOffset[N.getOpcode()]; } - + while (1) { assert(MatcherIndex < TableSize && "Invalid index"); #ifndef NDEBUG @@ -1915,7 +2124,7 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, // determine immediately that the first check (or first several) will // immediately fail, don't even bother pushing a scope for them. unsigned FailIndex; - + while (1) { unsigned NumToSkip = MatcherTable[MatcherIndex++]; if (NumToSkip & 128) @@ -1925,12 +2134,12 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, FailIndex = 0; break; } - + FailIndex = MatcherIndex+NumToSkip; - + unsigned MatcherIndexOfPredicate = MatcherIndex; (void)MatcherIndexOfPredicate; // silence warning. - + // If we can't evaluate this predicate without pushing a scope (e.g. if // it is a 'MoveParent') or if the predicate succeeds on this node, we // push the scope and evaluate the full predicate chain. @@ -1939,20 +2148,20 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, Result, *this, RecordedNodes); if (!Result) break; - + DEBUG(errs() << " Skipped scope entry (due to false predicate) at " << "index " << MatcherIndexOfPredicate << ", continuing at " << FailIndex << "\n"); ++NumDAGIselRetries; - + // Otherwise, we know that this case of the Scope is guaranteed to fail, // move to the next case. MatcherIndex = FailIndex; } - + // If the whole scope failed to match, bail. if (FailIndex == 0) break; - + // Push a MatchScope which indicates where to go if the first child fails // to match. MatchScope NewEntry; @@ -1961,17 +2170,21 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, NewEntry.NumRecordedNodes = RecordedNodes.size(); NewEntry.NumMatchedMemRefs = MatchedMemRefs.size(); NewEntry.InputChain = InputChain; - NewEntry.InputFlag = InputFlag; + NewEntry.InputGlue = InputGlue; NewEntry.HasChainNodesMatched = !ChainNodesMatched.empty(); - NewEntry.HasFlagResultNodesMatched = !FlagResultNodesMatched.empty(); + NewEntry.HasGlueResultNodesMatched = !GlueResultNodesMatched.empty(); MatchScopes.push_back(NewEntry); continue; } - case OPC_RecordNode: + case OPC_RecordNode: { // Remember this node, it may end up being an operand in the pattern. - RecordedNodes.push_back(N); + SDNode *Parent = 0; + if (NodeStack.size() > 1) + Parent = NodeStack[NodeStack.size()-2].getNode(); + RecordedNodes.push_back(std::make_pair(N, Parent)); continue; - + } + case OPC_RecordChild0: case OPC_RecordChild1: case OPC_RecordChild2: case OPC_RecordChild3: case OPC_RecordChild4: case OPC_RecordChild5: @@ -1980,20 +2193,21 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, if (ChildNo >= N.getNumOperands()) break; // Match fails if out of range child #. - RecordedNodes.push_back(N->getOperand(ChildNo)); + RecordedNodes.push_back(std::make_pair(N->getOperand(ChildNo), + N.getNode())); continue; } case OPC_RecordMemRef: MatchedMemRefs.push_back(cast(N)->getMemOperand()); continue; - - case OPC_CaptureFlagInput: - // If the current node has an input flag, capture it in InputFlag. + + case OPC_CaptureGlueInput: + // If the current node has an input glue, capture it in InputGlue. if (N->getNumOperands() != 0 && - N->getOperand(N->getNumOperands()-1).getValueType() == MVT::Flag) - InputFlag = N->getOperand(N->getNumOperands()-1); + N->getOperand(N->getNumOperands()-1).getValueType() == MVT::Glue) + InputGlue = N->getOperand(N->getNumOperands()-1); continue; - + case OPC_MoveChild: { unsigned ChildNo = MatcherTable[MatcherIndex++]; if (ChildNo >= N.getNumOperands()) @@ -2002,14 +2216,14 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, NodeStack.push_back(N); continue; } - + case OPC_MoveParent: // Pop the current node off the NodeStack. NodeStack.pop_back(); assert(!NodeStack.empty() && "Node stack imbalance!"); - N = NodeStack.back(); + N = NodeStack.back(); continue; - + case OPC_CheckSame: if (!::CheckSame(MatcherTable, MatcherIndex, N, RecordedNodes)) break; continue; @@ -2025,7 +2239,8 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, unsigned CPNum = MatcherTable[MatcherIndex++]; unsigned RecNo = MatcherTable[MatcherIndex++]; assert(RecNo < RecordedNodes.size() && "Invalid CheckComplexPat"); - if (!CheckComplexPattern(NodeToMatch, RecordedNodes[RecNo], CPNum, + if (!CheckComplexPattern(NodeToMatch, RecordedNodes[RecNo].second, + RecordedNodes[RecNo].first, CPNum, RecordedNodes)) break; continue; @@ -2033,11 +2248,11 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, case OPC_CheckOpcode: if (!::CheckOpcode(MatcherTable, MatcherIndex, N.getNode())) break; continue; - + case OPC_CheckType: if (!::CheckType(MatcherTable, MatcherIndex, N, TLI)) break; continue; - + case OPC_SwitchOpcode: { unsigned CurNodeOpcode = N.getOpcode(); unsigned SwitchStart = MatcherIndex-1; (void)SwitchStart; @@ -2055,22 +2270,22 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, // If the opcode matches, then we will execute this case. if (CurNodeOpcode == Opc) break; - + // Otherwise, skip over this case. MatcherIndex += CaseSize; } - + // If no cases matched, bail out. if (CaseSize == 0) break; - + // Otherwise, execute the case we found. DEBUG(errs() << " OpcodeSwitch from " << SwitchStart << " to " << MatcherIndex << "\n"); continue; } - + case OPC_SwitchType: { - MVT::SimpleValueType CurNodeVT = N.getValueType().getSimpleVT().SimpleTy; + MVT CurNodeVT = N.getValueType().getSimpleVT(); unsigned SwitchStart = MatcherIndex-1; (void)SwitchStart; unsigned CaseSize; while (1) { @@ -2079,23 +2294,22 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, if (CaseSize & 128) CaseSize = GetVBR(CaseSize, MatcherTable, MatcherIndex); if (CaseSize == 0) break; - - MVT::SimpleValueType CaseVT = - (MVT::SimpleValueType)MatcherTable[MatcherIndex++]; + + MVT CaseVT = (MVT::SimpleValueType)MatcherTable[MatcherIndex++]; if (CaseVT == MVT::iPTR) - CaseVT = TLI.getPointerTy().SimpleTy; - + CaseVT = TLI.getPointerTy(); + // If the VT matches, then we will execute this case. if (CurNodeVT == CaseVT) break; - + // Otherwise, skip over this case. MatcherIndex += CaseSize; } - + // If no cases matched, bail out. if (CaseSize == 0) break; - + // Otherwise, execute the case we found. DEBUG(errs() << " TypeSwitch[" << EVT(CurNodeVT).getEVTString() << "] from " << SwitchStart << " to " << MatcherIndex<<'\n'); @@ -2124,7 +2338,7 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, case OPC_CheckOrImm: if (!::CheckOrImm(MatcherTable, MatcherIndex, N, *this)) break; continue; - + case OPC_CheckFoldableChainNode: { assert(NodeStack.size() != 1 && "No parent node"); // Verify that all intermediate nodes between the root and this one have @@ -2145,7 +2359,7 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, NodeToMatch, OptLevel, true/*We validate our own chains*/)) break; - + continue; } case OPC_EmitInteger: { @@ -2154,22 +2368,36 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, int64_t Val = MatcherTable[MatcherIndex++]; if (Val & 128) Val = GetVBR(Val, MatcherTable, MatcherIndex); - RecordedNodes.push_back(CurDAG->getTargetConstant(Val, VT)); + RecordedNodes.push_back(std::pair( + CurDAG->getTargetConstant(Val, VT), (SDNode*)0)); continue; } case OPC_EmitRegister: { MVT::SimpleValueType VT = (MVT::SimpleValueType)MatcherTable[MatcherIndex++]; unsigned RegNo = MatcherTable[MatcherIndex++]; - RecordedNodes.push_back(CurDAG->getRegister(RegNo, VT)); + RecordedNodes.push_back(std::pair( + CurDAG->getRegister(RegNo, VT), (SDNode*)0)); + continue; + } + case OPC_EmitRegister2: { + // For targets w/ more than 256 register names, the register enum + // values are stored in two bytes in the matcher table (just like + // opcodes). + MVT::SimpleValueType VT = + (MVT::SimpleValueType)MatcherTable[MatcherIndex++]; + unsigned RegNo = MatcherTable[MatcherIndex++]; + RegNo |= MatcherTable[MatcherIndex++] << 8; + RecordedNodes.push_back(std::pair( + CurDAG->getRegister(RegNo, VT), (SDNode*)0)); continue; } - + case OPC_EmitConvertToTarget: { // Convert from IMM/FPIMM to target version. unsigned RecNo = MatcherTable[MatcherIndex++]; assert(RecNo < RecordedNodes.size() && "Invalid CheckSame"); - SDValue Imm = RecordedNodes[RecNo]; + SDValue Imm = RecordedNodes[RecNo].first; if (Imm->getOpcode() == ISD::Constant) { int64_t Val = cast(Imm)->getZExtValue(); @@ -2178,11 +2406,11 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, const ConstantFP *Val=cast(Imm)->getConstantFPValue(); Imm = CurDAG->getTargetConstantFP(*Val, Imm.getValueType()); } - - RecordedNodes.push_back(Imm); + + RecordedNodes.push_back(std::make_pair(Imm, RecordedNodes[RecNo].second)); continue; } - + case OPC_EmitMergeInputChains1_0: // OPC_EmitMergeInputChains, 1, 0 case OPC_EmitMergeInputChains1_1: { // OPC_EmitMergeInputChains, 1, 1 // These are space-optimized forms of OPC_EmitMergeInputChains. @@ -2190,28 +2418,28 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, "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"); - ChainNodesMatched.push_back(RecordedNodes[RecNo].getNode()); - + ChainNodesMatched.push_back(RecordedNodes[RecNo].first.getNode()); + // FIXME: What if other value results of the node have uses not matched // by this pattern? if (ChainNodesMatched.back() != NodeToMatch && - !RecordedNodes[RecNo].hasOneUse()) { + !RecordedNodes[RecNo].first.hasOneUse()) { ChainNodesMatched.clear(); break; } - + // Merge the input chains if they are not intra-pattern references. InputChain = HandleMergeInputChains(ChainNodesMatched, CurDAG); - + if (InputChain.getNode() == 0) break; // Failed to merge. continue; } - + case OPC_EmitMergeInputChains: { assert(InputChain.getNode() == 0 && "EmitMergeInputChains should be the first chain producing node"); @@ -2231,54 +2459,55 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, for (unsigned i = 0; i != NumChains; ++i) { unsigned RecNo = MatcherTable[MatcherIndex++]; assert(RecNo < RecordedNodes.size() && "Invalid CheckSame"); - ChainNodesMatched.push_back(RecordedNodes[RecNo].getNode()); - + ChainNodesMatched.push_back(RecordedNodes[RecNo].first.getNode()); + // FIXME: What if other value results of the node have uses not matched // by this pattern? if (ChainNodesMatched.back() != NodeToMatch && - !RecordedNodes[RecNo].hasOneUse()) { + !RecordedNodes[RecNo].first.hasOneUse()) { ChainNodesMatched.clear(); break; } } - + // If the inner loop broke out, the match fails. if (ChainNodesMatched.empty()) break; // Merge the input chains if they are not intra-pattern references. InputChain = HandleMergeInputChains(ChainNodesMatched, CurDAG); - + if (InputChain.getNode() == 0) break; // Failed to merge. continue; } - + case OPC_EmitCopyToReg: { unsigned RecNo = MatcherTable[MatcherIndex++]; assert(RecNo < RecordedNodes.size() && "Invalid CheckSame"); unsigned DestPhysReg = MatcherTable[MatcherIndex++]; - + if (InputChain.getNode() == 0) InputChain = CurDAG->getEntryNode(); - + InputChain = CurDAG->getCopyToReg(InputChain, NodeToMatch->getDebugLoc(), - DestPhysReg, RecordedNodes[RecNo], - InputFlag); - - InputFlag = InputChain.getValue(1); + DestPhysReg, RecordedNodes[RecNo].first, + InputGlue); + + InputGlue = InputChain.getValue(1); continue; } - + case OPC_EmitNodeXForm: { unsigned XFormNo = MatcherTable[MatcherIndex++]; unsigned RecNo = MatcherTable[MatcherIndex++]; assert(RecNo < RecordedNodes.size() && "Invalid CheckSame"); - RecordedNodes.push_back(RunSDNodeXForm(RecordedNodes[RecNo], XFormNo)); + SDValue Res = RunSDNodeXForm(RecordedNodes[RecNo].first, XFormNo); + RecordedNodes.push_back(std::pair(Res, (SDNode*) 0)); continue; } - + case OPC_EmitNode: case OPC_MorphNodeTo: { uint16_t TargetOpc = MatcherTable[MatcherIndex++]; @@ -2293,12 +2522,12 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, if (VT == MVT::iPTR) VT = TLI.getPointerTy().SimpleTy; VTs.push_back(VT); } - + if (EmitNodeInfo & OPFL_Chain) VTs.push_back(MVT::Other); - if (EmitNodeInfo & OPFL_FlagOutput) - VTs.push_back(MVT::Flag); - + if (EmitNodeInfo & OPFL_GlueOutput) + VTs.push_back(MVT::Glue); + // This is hot code, so optimize the two most common cases of 1 and 2 // results. SDVTList VTList; @@ -2316,11 +2545,11 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, unsigned RecNo = MatcherTable[MatcherIndex++]; if (RecNo & 128) RecNo = GetVBR(RecNo, MatcherTable, MatcherIndex); - + assert(RecNo < RecordedNodes.size() && "Invalid EmitNode"); - Ops.push_back(RecordedNodes[RecNo]); + Ops.push_back(RecordedNodes[RecNo].first); } - + // If there are variadic operands to add, handle them now. if (EmitNodeInfo & OPFL_VariadicInfo) { // Determine the start index to copy from. @@ -2328,22 +2557,22 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, FirstOpToCopy += (EmitNodeInfo & OPFL_Chain) ? 1 : 0; assert(NodeToMatch->getNumOperands() >= FirstOpToCopy && "Invalid variadic node"); - // Copy all of the variadic operands, not including a potential flag + // Copy all of the variadic operands, not including a potential glue // input. for (unsigned i = FirstOpToCopy, e = NodeToMatch->getNumOperands(); i != e; ++i) { SDValue V = NodeToMatch->getOperand(i); - if (V.getValueType() == MVT::Flag) break; + if (V.getValueType() == MVT::Glue) break; Ops.push_back(V); } } - - // If this has chain/flag inputs, add them. + + // If this has chain/glue inputs, add them. if (EmitNodeInfo & OPFL_Chain) Ops.push_back(InputChain); - if ((EmitNodeInfo & OPFL_FlagInput) && InputFlag.getNode() != 0) - Ops.push_back(InputFlag); - + if ((EmitNodeInfo & OPFL_GlueInput) && InputGlue.getNode() != 0) + Ops.push_back(InputGlue); + // Create the node. SDNode *Res = 0; if (Opcode != OPC_MorphNodeTo) { @@ -2351,71 +2580,106 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, // add the results to the RecordedNodes list. Res = CurDAG->getMachineNode(TargetOpc, NodeToMatch->getDebugLoc(), VTList, Ops.data(), Ops.size()); - - // Add all the non-flag/non-chain results to the RecordedNodes list. + + // 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::Flag) break; - RecordedNodes.push_back(SDValue(Res, i)); + if (VTs[i] == MVT::Other || VTs[i] == MVT::Glue) break; + RecordedNodes.push_back(std::pair(SDValue(Res, i), + (SDNode*) 0)); } - + } else { Res = MorphNode(NodeToMatch, TargetOpc, VTList, Ops.data(), Ops.size(), EmitNodeInfo); } - - // If the node had chain/flag results, update our notion of the current - // chain and flag. - if (EmitNodeInfo & OPFL_FlagOutput) { - InputFlag = SDValue(Res, VTs.size()-1); + + // If the node had chain/glue results, update our notion of the current + // chain and glue. + if (EmitNodeInfo & OPFL_GlueOutput) { + InputGlue = SDValue(Res, VTs.size()-1); if (EmitNodeInfo & OPFL_Chain) InputChain = SDValue(Res, VTs.size()-2); } else if (EmitNodeInfo & OPFL_Chain) InputChain = SDValue(Res, VTs.size()-1); - // If the OPFL_MemRefs flag is set on this node, slap all of the + // If the OPFL_MemRefs glue is set on this node, slap all of the // accumulated memrefs onto it. // // FIXME: This is vastly incorrect for patterns with multiple outputs // instructions that access memory and for ComplexPatterns that match // loads. if (EmitNodeInfo & OPFL_MemRefs) { + // Only attach load or store memory operands if the generated + // instruction may load or store. + const TargetInstrDesc &TID = TM.getInstrInfo()->get(TargetOpc); + bool mayLoad = TID.mayLoad(); + bool mayStore = TID.mayStore(); + + unsigned NumMemRefs = 0; + for (SmallVector::const_iterator I = + MatchedMemRefs.begin(), E = MatchedMemRefs.end(); I != E; ++I) { + if ((*I)->isLoad()) { + if (mayLoad) + ++NumMemRefs; + } else if ((*I)->isStore()) { + if (mayStore) + ++NumMemRefs; + } else { + ++NumMemRefs; + } + } + MachineSDNode::mmo_iterator MemRefs = - MF->allocateMemRefsArray(MatchedMemRefs.size()); - std::copy(MatchedMemRefs.begin(), MatchedMemRefs.end(), MemRefs); + MF->allocateMemRefsArray(NumMemRefs); + + MachineSDNode::mmo_iterator MemRefsPos = MemRefs; + for (SmallVector::const_iterator I = + MatchedMemRefs.begin(), E = MatchedMemRefs.end(); I != E; ++I) { + if ((*I)->isLoad()) { + if (mayLoad) + *MemRefsPos++ = *I; + } else if ((*I)->isStore()) { + if (mayStore) + *MemRefsPos++ = *I; + } else { + *MemRefsPos++ = *I; + } + } + cast(Res) - ->setMemRefs(MemRefs, MemRefs + MatchedMemRefs.size()); + ->setMemRefs(MemRefs, MemRefs + NumMemRefs); } - + DEBUG(errs() << " " << (Opcode == OPC_MorphNodeTo ? "Morphed" : "Created") << " node: "; Res->dump(CurDAG); errs() << "\n"); - + // If this was a MorphNodeTo then we're completely done! if (Opcode == OPC_MorphNodeTo) { - // Update chain and flag uses. - UpdateChainsAndFlags(NodeToMatch, InputChain, ChainNodesMatched, - InputFlag, FlagResultNodesMatched, true); + // Update chain and glue uses. + UpdateChainsAndGlue(NodeToMatch, InputChain, ChainNodesMatched, + InputGlue, GlueResultNodesMatched, true); return Res; } - + continue; } - - case OPC_MarkFlagResults: { + + case OPC_MarkGlueResults: { unsigned NumNodes = MatcherTable[MatcherIndex++]; - - // Read and remember all the flag-result nodes. + + // Read and remember all the glue-result nodes. for (unsigned i = 0; i != NumNodes; ++i) { unsigned RecNo = MatcherTable[MatcherIndex++]; if (RecNo & 128) RecNo = GetVBR(RecNo, MatcherTable, MatcherIndex); assert(RecNo < RecordedNodes.size() && "Invalid CheckSame"); - FlagResultNodesMatched.push_back(RecordedNodes[RecNo].getNode()); + GlueResultNodesMatched.push_back(RecordedNodes[RecNo].first.getNode()); } continue; } - + case OPC_CompleteMatch: { // The match has been completed, and any new nodes (if any) have been // created. Patch up references to the matched dag to use the newly @@ -2426,13 +2690,13 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, unsigned ResSlot = MatcherTable[MatcherIndex++]; if (ResSlot & 128) ResSlot = GetVBR(ResSlot, MatcherTable, MatcherIndex); - + assert(ResSlot < RecordedNodes.size() && "Invalid CheckSame"); - SDValue Res = RecordedNodes[ResSlot]; - + SDValue Res = RecordedNodes[ResSlot].first; + assert(i < NodeToMatch->getNumValues() && NodeToMatch->getValueType(i) != MVT::Other && - NodeToMatch->getValueType(i) != MVT::Flag && + NodeToMatch->getValueType(i) != MVT::Glue && "Invalid number of results to complete!"); assert((NodeToMatch->getValueType(i) == Res.getValueType() || NodeToMatch->getValueType(i) == MVT::iPTR || @@ -2443,24 +2707,23 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, CurDAG->ReplaceAllUsesOfValueWith(SDValue(NodeToMatch, i), Res); } - // If the root node defines a flag, add it to the flag nodes to update - // list. - if (NodeToMatch->getValueType(NodeToMatch->getNumValues()-1) == MVT::Flag) - FlagResultNodesMatched.push_back(NodeToMatch); - - // Update chain and flag uses. - UpdateChainsAndFlags(NodeToMatch, InputChain, ChainNodesMatched, - InputFlag, FlagResultNodesMatched, false); - + // If the root node defines glue, add it to the glue nodes to update list. + if (NodeToMatch->getValueType(NodeToMatch->getNumValues()-1) == MVT::Glue) + GlueResultNodesMatched.push_back(NodeToMatch); + + // Update chain and glue uses. + UpdateChainsAndGlue(NodeToMatch, InputChain, ChainNodesMatched, + InputGlue, GlueResultNodesMatched, false); + assert(NodeToMatch->use_empty() && "Didn't replace all uses of the node?"); - + // FIXME: We just return here, which interacts correctly with SelectRoot // above. We should fix this to not return an SDNode* anymore. return 0; } } - + // 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. @@ -2483,15 +2746,15 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, if (LastScope.NumMatchedMemRefs != MatchedMemRefs.size()) MatchedMemRefs.resize(LastScope.NumMatchedMemRefs); MatcherIndex = LastScope.FailIndex; - + DEBUG(errs() << " Continuing at " << MatcherIndex << "\n"); - + InputChain = LastScope.InputChain; - InputFlag = LastScope.InputFlag; + InputGlue = LastScope.InputGlue; if (!LastScope.HasChainNodesMatched) ChainNodesMatched.clear(); - if (!LastScope.HasFlagResultNodesMatched) - FlagResultNodesMatched.clear(); + if (!LastScope.HasGlueResultNodesMatched) + GlueResultNodesMatched.clear(); // Check to see what the offset is at the new MatcherIndex. If it is zero // we have reached the end of this scope, otherwise we have another child @@ -2506,21 +2769,21 @@ SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, LastScope.FailIndex = MatcherIndex+NumToSkip; break; } - + // End of this scope, pop it and try the next child in the containing // scope. MatchScopes.pop_back(); } } } - + void SelectionDAGISel::CannotYetSelect(SDNode *N) { std::string msg; raw_string_ostream Msg(msg); - Msg << "Cannot yet select: "; - + Msg << "Cannot select: "; + if (N->getOpcode() != ISD::INTRINSIC_W_CHAIN && N->getOpcode() != ISD::INTRINSIC_WO_CHAIN && N->getOpcode() != ISD::INTRINSIC_VOID) {