diff options
Diffstat (limited to 'lib/CodeGen')
42 files changed, 2129 insertions, 555 deletions
diff --git a/lib/CodeGen/AsmPrinter/AsmPrinter.cpp b/lib/CodeGen/AsmPrinter/AsmPrinter.cpp index fc08384..bd2b1b6 100644 --- a/lib/CodeGen/AsmPrinter/AsmPrinter.cpp +++ b/lib/CodeGen/AsmPrinter/AsmPrinter.cpp @@ -917,11 +917,10 @@ void AsmPrinter::EmitAlignment(unsigned NumBits, const GlobalValue *GV, if (NumBits == 0) return; // No need to emit alignment. - unsigned FillValue = 0; if (getCurrentSection()->getKind().isText()) - FillValue = MAI->getTextAlignFillValue(); - - OutStreamer.EmitValueToAlignment(1 << NumBits, FillValue, 1, 0); + OutStreamer.EmitCodeAlignment(1 << NumBits); + else + OutStreamer.EmitValueToAlignment(1 << NumBits, 0, 1, 0); } /// LowerConstant - Lower the specified LLVM Constant to an MCExpr. @@ -1717,7 +1716,7 @@ void AsmPrinter::EmitBasicBlockStart(const MachineBasicBlock *MBB) const { } // Print the main label for the block. - if (MBB->pred_empty() || MBB->isOnlyReachableByFallthrough()) { + if (MBB->pred_empty() || isBlockOnlyReachableByFallthrough(MBB)) { if (VerboseAsm) { // NOTE: Want this comment at start of line. O << MAI->getCommentString() << " BB#" << MBB->getNumber() << ':'; @@ -1764,6 +1763,39 @@ void AsmPrinter::printOffset(int64_t Offset) const { O << Offset; } +/// isBlockOnlyReachableByFallthough - Return true if the basic block has +/// exactly one predecessor and the control transfer mechanism between +/// the predecessor and this block is a fall-through. +bool AsmPrinter::isBlockOnlyReachableByFallthrough(const MachineBasicBlock *MBB) + const { + // If this is a landing pad, it isn't a fall through. If it has no preds, + // then nothing falls through to it. + if (MBB->isLandingPad() || MBB->pred_empty()) + return false; + + // If there isn't exactly one predecessor, it can't be a fall through. + MachineBasicBlock::const_pred_iterator PI = MBB->pred_begin(), PI2 = PI; + ++PI2; + if (PI2 != MBB->pred_end()) + return false; + + // The predecessor has to be immediately before this block. + const MachineBasicBlock *Pred = *PI; + + if (!Pred->isLayoutSuccessor(MBB)) + return false; + + // If the block is completely empty, then it definitely does fall through. + if (Pred->empty()) + return true; + + // Otherwise, check the last instruction. + const MachineInstr &LastInst = Pred->back(); + return !LastInst.getDesc().isBarrier(); +} + + + GCMetadataPrinter *AsmPrinter::GetOrCreateGCPrinter(GCStrategy *S) { if (!S->usesMetadata()) return 0; diff --git a/lib/CodeGen/AsmPrinter/DwarfDebug.cpp b/lib/CodeGen/AsmPrinter/DwarfDebug.cpp index 5093dd9..5ad1e5e 100644 --- a/lib/CodeGen/AsmPrinter/DwarfDebug.cpp +++ b/lib/CodeGen/AsmPrinter/DwarfDebug.cpp @@ -238,7 +238,18 @@ public: LIndex = DSI; } } - setLastInsn(LastInsn); + + unsigned CurrentLastInsnIndex = 0; + if (const MachineInstr *CL = getLastInsn()) + CurrentLastInsnIndex = MIIndexMap[CL]; + unsigned FIndex = MIIndexMap[getFirstInsn()]; + + // Set LastInsn as the last instruction for this scope only if + // it follows + // 1) this scope's first instruction and + // 2) current last instruction for this scope, if any. + if (LIndex >= CurrentLastInsnIndex && LIndex >= FIndex) + setLastInsn(LastInsn); } #ifndef NDEBUG @@ -1166,7 +1177,9 @@ DIE *DwarfDebug::createSubprogramDIE(const DISubprogram &SP, bool MakeDecl) { return SPDie; SPDie = new DIE(dwarf::DW_TAG_subprogram); - addString(SPDie, dwarf::DW_AT_name, dwarf::DW_FORM_string, SP.getName()); + // Constructors and operators for anonymous aggregates do not have names. + if (!SP.getName().empty()) + addString(SPDie, dwarf::DW_AT_name, dwarf::DW_FORM_string, SP.getName()); StringRef LinkageName = SP.getLinkageName(); if (!LinkageName.empty()) diff --git a/lib/CodeGen/AsmPrinter/DwarfException.cpp b/lib/CodeGen/AsmPrinter/DwarfException.cpp index c6c59f5..2b08ba4 100644 --- a/lib/CodeGen/AsmPrinter/DwarfException.cpp +++ b/lib/CodeGen/AsmPrinter/DwarfException.cpp @@ -638,18 +638,18 @@ void DwarfException::EmitExceptionTable() { const unsigned LandingPadSize = SizeOfEncodedValue(dwarf::DW_EH_PE_udata4); bool IsSJLJ = MAI->getExceptionHandlingType() == ExceptionHandling::SjLj; bool HaveTTData = IsSJLJ ? (!TypeInfos.empty() || !FilterIds.empty()) : true; - unsigned SizeSites; + unsigned CallSiteTableLength; if (IsSJLJ) - SizeSites = 0; + CallSiteTableLength = 0; else - SizeSites = CallSites.size() * + CallSiteTableLength = CallSites.size() * (SiteStartSize + SiteLengthSize + LandingPadSize); for (unsigned i = 0, e = CallSites.size(); i < e; ++i) { - SizeSites += MCAsmInfo::getULEB128Size(CallSites[i].Action); + CallSiteTableLength += MCAsmInfo::getULEB128Size(CallSites[i].Action); if (IsSJLJ) - SizeSites += MCAsmInfo::getULEB128Size(i); + CallSiteTableLength += MCAsmInfo::getULEB128Size(i); } // Type infos. @@ -698,7 +698,20 @@ void DwarfException::EmitExceptionTable() { Asm->OutStreamer.SwitchSection(LSDASection); Asm->EmitAlignment(2, 0, 0, false); + // Emit the LSDA. O << "GCC_except_table" << SubprogramCount << ":\n"; + EmitLabel("exception", SubprogramCount); + + if (IsSJLJ) { + SmallString<16> LSDAName; + raw_svector_ostream(LSDAName) << MAI->getPrivateGlobalPrefix() << + "_LSDA_" << Asm->getFunctionNumber(); + O << LSDAName.str() << ":\n"; + } + + // Emit the LSDA header. + EmitEncodingByte(dwarf::DW_EH_PE_omit, "@LPStart"); + EmitEncodingByte(TTypeEncoding, "@TType"); // The type infos need to be aligned. GCC does this by inserting padding just // before the type infos. However, this changes the size of the exception @@ -707,7 +720,7 @@ void DwarfException::EmitExceptionTable() { // So by increasing the size by inserting padding, you may increase the number // of bytes used for writing the size. If it increases, say by one byte, then // you now need to output one less byte of padding to get the type infos - // aligned. However this decreases the size of the exception table. This + // aligned. However this decreases the size of the exception table. This // changes the value you have to output for the exception table size. Due to // the variable length encoding, the number of bytes used for writing the // length may decrease. If so, you then have to increase the amount of @@ -716,41 +729,35 @@ void DwarfException::EmitExceptionTable() { // We chose another solution: don't output padding inside the table like GCC // does, instead output it before the table. unsigned SizeTypes = TypeInfos.size() * TypeFormatSize; - unsigned TyOffset = sizeof(int8_t) + // Call site format - MCAsmInfo::getULEB128Size(SizeSites) + // Call site table length - SizeSites + SizeActions + SizeTypes; - unsigned TotalSize = sizeof(int8_t) + // LPStart format - sizeof(int8_t) + // TType format - (HaveTTData ? - MCAsmInfo::getULEB128Size(TyOffset) : 0) + // TType base offset - TyOffset; + unsigned CallSiteTableLengthSize = + MCAsmInfo::getULEB128Size(CallSiteTableLength); + unsigned TTypeBaseOffset = + sizeof(int8_t) + // Call site format + CallSiteTableLengthSize + // Call site table length size + CallSiteTableLength + // Call site table length + SizeActions + // Actions size + SizeTypes; + unsigned TTypeBaseOffsetSize = MCAsmInfo::getULEB128Size(TTypeBaseOffset); + unsigned TotalSize = + sizeof(int8_t) + // LPStart format + sizeof(int8_t) + // TType format + (HaveTTData ? TTypeBaseOffsetSize : 0) + // TType base offset size + TTypeBaseOffset; // TType base offset unsigned SizeAlign = (4 - TotalSize) & 3; - for (unsigned i = 0; i != SizeAlign; ++i) { - Asm->EmitInt8(0); - EOL("Padding"); - } - - EmitLabel("exception", SubprogramCount); - - if (IsSJLJ) { - SmallString<16> LSDAName; - raw_svector_ostream(LSDAName) << MAI->getPrivateGlobalPrefix() << - "_LSDA_" << Asm->getFunctionNumber(); - O << LSDAName.str() << ":\n"; + if (HaveTTData) { + // Account for any extra padding that will be added to the call site table + // length. + EmitULEB128(TTypeBaseOffset, "@TType base offset", SizeAlign); + SizeAlign = 0; } - // Emit the header. - EmitEncodingByte(dwarf::DW_EH_PE_omit, "@LPStart"); - EmitEncodingByte(TTypeEncoding, "@TType"); - - if (HaveTTData) - EmitULEB128(TyOffset, "@TType base offset"); - // SjLj Exception handling if (IsSJLJ) { EmitEncodingByte(dwarf::DW_EH_PE_udata4, "Call site"); - EmitULEB128(SizeSites, "Call site table length"); + + // Add extra padding if it wasn't added to the TType base offset. + EmitULEB128(CallSiteTableLength, "Call site table length", SizeAlign); // Emit the landing pad site information. unsigned idx = 0; @@ -791,7 +798,9 @@ void DwarfException::EmitExceptionTable() { // Emit the landing pad call site table. EmitEncodingByte(dwarf::DW_EH_PE_udata4, "Call site"); - EmitULEB128(SizeSites, "Call site table length"); + + // Add extra padding if it wasn't added to the TType base offset. + EmitULEB128(CallSiteTableLength, "Call site table length", SizeAlign); for (SmallVectorImpl<CallSiteEntry>::const_iterator I = CallSites.begin(), E = CallSites.end(); I != E; ++I) { diff --git a/lib/CodeGen/AsmPrinter/DwarfPrinter.cpp b/lib/CodeGen/AsmPrinter/DwarfPrinter.cpp index 1299d04..28ff0eb 100644 --- a/lib/CodeGen/AsmPrinter/DwarfPrinter.cpp +++ b/lib/CodeGen/AsmPrinter/DwarfPrinter.cpp @@ -159,29 +159,35 @@ void DwarfPrinter::EmitSLEB128(int Value, const char *Desc) const { Value >>= 7; IsMore = Value != Sign || ((Byte ^ Sign) & 0x40) != 0; if (IsMore) Byte |= 0x80; - Asm->OutStreamer.EmitIntValue(Byte, 1, /*addrspace*/0); } while (IsMore); } /// EmitULEB128 - emit the specified signed leb128 value. -void DwarfPrinter::EmitULEB128(unsigned Value, const char *Desc) const { +void DwarfPrinter::EmitULEB128(unsigned Value, const char *Desc, + unsigned PadTo) const { if (Asm->VerboseAsm && Desc) Asm->OutStreamer.AddComment(Desc); - if (MAI->hasLEB128()) { + if (MAI->hasLEB128() && PadTo == 0) { O << "\t.uleb128\t" << Value; Asm->OutStreamer.AddBlankLine(); return; } - // If we don't have .uleb128, emit as .bytes. + // If we don't have .uleb128 or we want to emit padding, emit as .bytes. do { unsigned char Byte = static_cast<unsigned char>(Value & 0x7f); Value >>= 7; - if (Value) Byte |= 0x80; + if (Value || PadTo != 0) Byte |= 0x80; Asm->OutStreamer.EmitIntValue(Byte, 1, /*addrspace*/0); } while (Value); + + if (PadTo) { + if (PadTo > 1) + Asm->OutStreamer.EmitFill(PadTo - 1, 0x80/*fillval*/, 0/*addrspace*/); + Asm->OutStreamer.EmitFill(1, 0/*fillval*/, 0/*addrspace*/); + } } diff --git a/lib/CodeGen/AsmPrinter/DwarfPrinter.h b/lib/CodeGen/AsmPrinter/DwarfPrinter.h index 73de398..bd715f2 100644 --- a/lib/CodeGen/AsmPrinter/DwarfPrinter.h +++ b/lib/CodeGen/AsmPrinter/DwarfPrinter.h @@ -111,7 +111,8 @@ public: void EmitSLEB128(int Value, const char *Desc) const; /// EmitULEB128 - emit the specified unsigned leb128 value. - void EmitULEB128(unsigned Value, const char *Desc = 0) const; + void EmitULEB128(unsigned Value, const char *Desc = 0, + unsigned PadTo = 0) const; /// PrintLabelName - Print label name in form used by Dwarf writer. diff --git a/lib/CodeGen/CMakeLists.txt b/lib/CodeGen/CMakeLists.txt index 62cf339..d385b86 100644 --- a/lib/CodeGen/CMakeLists.txt +++ b/lib/CodeGen/CMakeLists.txt @@ -22,6 +22,7 @@ add_llvm_library(LLVMCodeGen LiveVariables.cpp LowerSubregs.cpp MachineBasicBlock.cpp + MachineCSE.cpp MachineDominators.cpp MachineFunction.cpp MachineFunctionAnalysis.cpp diff --git a/lib/CodeGen/CalcSpillWeights.cpp b/lib/CodeGen/CalcSpillWeights.cpp index 2bedd04..a328d0e 100644 --- a/lib/CodeGen/CalcSpillWeights.cpp +++ b/lib/CodeGen/CalcSpillWeights.cpp @@ -131,10 +131,7 @@ bool CalculateSpillWeights::runOnMachineFunction(MachineFunction &fn) { if (Hint.first || Hint.second) li.weight *= 1.01F; - // Divide the weight of the interval by its size. This encourages - // spilling of intervals that are large and have few uses, and - // discourages spilling of small intervals with many uses. - li.weight /= lis->getApproximateInstructionCount(li) * SlotIndex::NUM; + lis->normalizeSpillWeight(li); } } diff --git a/lib/CodeGen/CodePlacementOpt.cpp b/lib/CodeGen/CodePlacementOpt.cpp index a13a310..3ff2a04 100644 --- a/lib/CodeGen/CodePlacementOpt.cpp +++ b/lib/CodeGen/CodePlacementOpt.cpp @@ -102,13 +102,6 @@ bool CodePlacementOpt::HasAnalyzableTerminator(MachineBasicBlock *MBB) { // Conservatively ignore EH landing pads. if (MBB->isLandingPad()) return false; - // Ignore blocks which look like they might have EH-related control flow. - // At the time of this writing, there are blocks which AnalyzeBranch - // thinks end in single uncoditional branches, yet which have two CFG - // successors. Code in this file is not prepared to reason about such things. - if (!MBB->empty() && MBB->back().isEHLabel()) - return false; - // Aggressively handle return blocks and similar constructs. if (MBB->succ_empty()) return true; @@ -118,6 +111,14 @@ bool CodePlacementOpt::HasAnalyzableTerminator(MachineBasicBlock *MBB) { // Make sure the terminator is understood. if (TII->AnalyzeBranch(*MBB, TBB, FBB, Cond)) return false; + // Ignore blocks which look like they might have EH-related control flow. + // AnalyzeBranch thinks it knows how to analyze such things, but it doesn't + // recognize the possibility of a control transfer through an unwind. + // Such blocks contain EH_LABEL instructions, however they may be in the + // middle of the block. Instead of searching for them, just check to see + // if the CFG disagrees with AnalyzeBranch. + if (1u + !Cond.empty() != MBB->succ_size()) + return false; // Make sure we have the option of reversing the condition. if (!Cond.empty() && TII->ReverseBranchCondition(Cond)) return false; diff --git a/lib/CodeGen/DeadMachineInstructionElim.cpp b/lib/CodeGen/DeadMachineInstructionElim.cpp index a215a19..d69c995 100644 --- a/lib/CodeGen/DeadMachineInstructionElim.cpp +++ b/lib/CodeGen/DeadMachineInstructionElim.cpp @@ -55,7 +55,7 @@ FunctionPass *llvm::createDeadMachineInstructionElimPass() { bool DeadMachineInstructionElim::isDead(const MachineInstr *MI) const { // Don't delete instructions with side effects. bool SawStore = false; - if (!MI->isSafeToMove(TII, SawStore, 0) && !MI->isPHI()) + if (!MI->isSafeToMove(TII, 0, SawStore) && !MI->isPHI()) return false; // Examine each operand. diff --git a/lib/CodeGen/LLVMTargetMachine.cpp b/lib/CodeGen/LLVMTargetMachine.cpp index 278de02..fd442db 100644 --- a/lib/CodeGen/LLVMTargetMachine.cpp +++ b/lib/CodeGen/LLVMTargetMachine.cpp @@ -67,6 +67,9 @@ static cl::opt<bool> VerifyMachineCode("verify-machineinstrs", cl::Hidden, cl::desc("Verify generated machine code"), cl::init(getenv("LLVM_VERIFY_MACHINEINSTRS")!=NULL)); +static cl::opt<bool> EnableMachineCSE("machine-cse", cl::Hidden, + cl::desc("Enable Machine CSE")); + static cl::opt<cl::boolOrDefault> AsmVerbose("asm-verbose", cl::desc("Add comments to directives."), cl::init(cl::BOU_UNSET)); @@ -115,9 +118,10 @@ LLVMTargetMachine::setCodeModelForStatic() { bool LLVMTargetMachine::addPassesToEmitFile(PassManagerBase &PM, formatted_raw_ostream &Out, CodeGenFileType FileType, - CodeGenOpt::Level OptLevel) { + CodeGenOpt::Level OptLevel, + bool DisableVerify) { // Add common CodeGen passes. - if (addCommonCodeGenPasses(PM, OptLevel)) + if (addCommonCodeGenPasses(PM, OptLevel, DisableVerify)) return true; OwningPtr<MCContext> Context(new MCContext()); @@ -193,12 +197,13 @@ bool LLVMTargetMachine::addPassesToEmitFile(PassManagerBase &PM, /// bool LLVMTargetMachine::addPassesToEmitMachineCode(PassManagerBase &PM, JITCodeEmitter &JCE, - CodeGenOpt::Level OptLevel) { + CodeGenOpt::Level OptLevel, + bool DisableVerify) { // Make sure the code model is set. setCodeModelForJIT(); // Add common CodeGen passes. - if (addCommonCodeGenPasses(PM, OptLevel)) + if (addCommonCodeGenPasses(PM, OptLevel, DisableVerify)) return true; addCodeEmitter(PM, OptLevel, JCE); @@ -221,13 +226,19 @@ static void printAndVerify(PassManagerBase &PM, /// emitting to assembly files or machine code output. /// bool LLVMTargetMachine::addCommonCodeGenPasses(PassManagerBase &PM, - CodeGenOpt::Level OptLevel) { + CodeGenOpt::Level OptLevel, + bool DisableVerify) { // Standard LLVM-Level Passes. + // Before running any passes, run the verifier to determine if the input + // coming from the front-end and/or optimizer is valid. + if (!DisableVerify) + PM.add(createVerifierPass()); + // Optionally, tun split-GEPs and no-load GVN. if (EnableSplitGEPGVN) { PM.add(createGEPSplitterPass()); - PM.add(createGVNPass(/*NoPRE=*/false, /*NoLoads=*/true)); + PM.add(createGVNPass(/*NoLoads=*/true)); } // Run loop strength reduction before anything else. @@ -235,9 +246,6 @@ bool LLVMTargetMachine::addCommonCodeGenPasses(PassManagerBase &PM, PM.add(createLoopStrengthReducePass(getTargetLowering())); if (PrintLSR) PM.add(createPrintFunctionPass("\n\n*** Code after LSR ***\n", &dbgs())); -#ifndef NDEBUG - PM.add(createVerifierPass()); -#endif } // Turn exception handling constructs into something the code generators can @@ -277,6 +285,11 @@ bool LLVMTargetMachine::addCommonCodeGenPasses(PassManagerBase &PM, "*** Final LLVM Code input to ISel ***\n", &dbgs())); + // All passes which modify the LLVM IR are now complete; run the verifier + // to ensure that the IR is valid. + if (!DisableVerify) + PM.add(createVerifierPass()); + // Standard Lower-Level Passes. // Set up a MachineFunction for the rest of CodeGen to work on. @@ -307,6 +320,8 @@ bool LLVMTargetMachine::addCommonCodeGenPasses(PassManagerBase &PM, if (OptLevel != CodeGenOpt::None) { PM.add(createOptimizeExtsPass()); + if (EnableMachineCSE) + PM.add(createMachineCSEPass()); if (!DisableMachineLICM) PM.add(createMachineLICMPass()); if (!DisableMachineSink) diff --git a/lib/CodeGen/LiveIntervalAnalysis.cpp b/lib/CodeGen/LiveIntervalAnalysis.cpp index f6bf433..ccda66f 100644 --- a/lib/CodeGen/LiveIntervalAnalysis.cpp +++ b/lib/CodeGen/LiveIntervalAnalysis.cpp @@ -329,24 +329,43 @@ void LiveIntervals::handleVirtualRegisterDef(MachineBasicBlock *mbb, DEBUG(dbgs() << " +" << NewLR); interval.addRange(NewLR); - // Iterate over all of the blocks that the variable is completely - // live in, adding [insrtIndex(begin), instrIndex(end)+4) to the - // live interval. - for (SparseBitVector<>::iterator I = vi.AliveBlocks.begin(), - E = vi.AliveBlocks.end(); I != E; ++I) { - MachineBasicBlock *aliveBlock = mf_->getBlockNumbered(*I); - LiveRange LR(getMBBStartIdx(aliveBlock), getMBBEndIdx(aliveBlock), ValNo); - interval.addRange(LR); - DEBUG(dbgs() << " +" << LR); + bool PHIJoin = lv_->isPHIJoin(interval.reg); + + if (PHIJoin) { + // A phi join register is killed at the end of the MBB and revived as a new + // valno in the killing blocks. + assert(vi.AliveBlocks.empty() && "Phi join can't pass through blocks"); + DEBUG(dbgs() << " phi-join"); + ValNo->addKill(indexes_->getTerminatorGap(mbb)); + ValNo->setHasPHIKill(true); + } else { + // Iterate over all of the blocks that the variable is completely + // live in, adding [insrtIndex(begin), instrIndex(end)+4) to the + // live interval. + for (SparseBitVector<>::iterator I = vi.AliveBlocks.begin(), + E = vi.AliveBlocks.end(); I != E; ++I) { + MachineBasicBlock *aliveBlock = mf_->getBlockNumbered(*I); + LiveRange LR(getMBBStartIdx(aliveBlock), getMBBEndIdx(aliveBlock), ValNo); + interval.addRange(LR); + DEBUG(dbgs() << " +" << LR); + } } // Finally, this virtual register is live from the start of any killing // block to the 'use' slot of the killing instruction. for (unsigned i = 0, e = vi.Kills.size(); i != e; ++i) { MachineInstr *Kill = vi.Kills[i]; - SlotIndex killIdx = - getInstructionIndex(Kill).getDefIndex(); - LiveRange LR(getMBBStartIdx(Kill->getParent()), killIdx, ValNo); + SlotIndex Start = getMBBStartIdx(Kill->getParent()); + SlotIndex killIdx = getInstructionIndex(Kill).getDefIndex(); + + // Create interval with one of a NEW value number. Note that this value + // number isn't actually defined by an instruction, weird huh? :) + if (PHIJoin) { + ValNo = interval.getNextValue(SlotIndex(Start, true), 0, false, + VNInfoAllocator); + ValNo->setIsPHIDef(true); + } + LiveRange LR(Start, killIdx, ValNo); interval.addRange(LR); ValNo->addKill(killIdx); DEBUG(dbgs() << " +" << LR); @@ -409,48 +428,11 @@ void LiveIntervals::handleVirtualRegisterDef(MachineBasicBlock *mbb, interval.print(dbgs(), tri_); }); } else { - // Otherwise, this must be because of phi elimination. If this is the - // first redefinition of the vreg that we have seen, go back and change - // the live range in the PHI block to be a different value number. - if (interval.containsOneValue()) { - - VNInfo *VNI = interval.getValNumInfo(0); - // Phi elimination may have reused the register for multiple identical - // phi nodes. There will be a kill per phi. Remove the old ranges that - // we now know have an incorrect number. - for (unsigned ki=0, ke=vi.Kills.size(); ki != ke; ++ki) { - MachineInstr *Killer = vi.Kills[ki]; - SlotIndex Start = getMBBStartIdx(Killer->getParent()); - SlotIndex End = getInstructionIndex(Killer).getDefIndex(); - DEBUG({ - dbgs() << "\n\t\trenaming [" << Start << "," << End << "] in: "; - interval.print(dbgs(), tri_); - }); - interval.removeRange(Start, End); - - // Replace the interval with one of a NEW value number. Note that - // this value number isn't actually defined by an instruction, weird - // huh? :) - LiveRange LR(Start, End, - interval.getNextValue(SlotIndex(Start, true), - 0, false, VNInfoAllocator)); - LR.valno->setIsPHIDef(true); - interval.addRange(LR); - LR.valno->addKill(End); - } - - MachineBasicBlock *killMBB = getMBBFromIndex(VNI->def); - VNI->addKill(indexes_->getTerminatorGap(killMBB)); - VNI->setHasPHIKill(true); - DEBUG({ - dbgs() << " RESULT: "; - interval.print(dbgs(), tri_); - }); - } - + assert(lv_->isPHIJoin(interval.reg) && "Multiply defined register"); // In the case of PHI elimination, each variable definition is only // live until the end of the block. We've already taken care of the // rest of the live range. + SlotIndex defIndex = MIIdx.getDefIndex(); if (MO.isEarlyClobber()) defIndex = MIIdx.getUseIndex(); @@ -468,7 +450,7 @@ void LiveIntervals::handleVirtualRegisterDef(MachineBasicBlock *mbb, interval.addRange(LR); ValNo->addKill(indexes_->getTerminatorGap(mbb)); ValNo->setHasPHIKill(true); - DEBUG(dbgs() << " +" << LR); + DEBUG(dbgs() << " phi-join +" << LR); } } @@ -1358,11 +1340,9 @@ rewriteInstructionsForSpills(const LiveInterval &li, bool TrySplit, MachineBasicBlock *MBB = MI->getParent(); if (ImpUse && MI != ReMatDefMI) { - // Re-matting an instruction with virtual register use. Update the - // register interval's spill weight to HUGE_VALF to prevent it from - // being spilled. - LiveInterval &ImpLi = getInterval(ImpUse); - ImpLi.weight = HUGE_VALF; + // Re-matting an instruction with virtual register use. Prevent interval + // from being spilled. + getInterval(ImpUse).markNotSpillable(); } unsigned MBBId = MBB->getNumber(); @@ -1414,7 +1394,7 @@ rewriteInstructionsForSpills(const LiveInterval &li, bool TrySplit, LiveInterval &nI = getOrCreateInterval(NewVReg); if (!TrySplit) { // The spill weight is now infinity as it cannot be spilled again. - nI.weight = HUGE_VALF; + nI.markNotSpillable(); continue; } @@ -1562,6 +1542,28 @@ LiveIntervals::handleSpilledImpDefs(const LiveInterval &li, VirtRegMap &vrm, } } +float +LiveIntervals::getSpillWeight(bool isDef, bool isUse, unsigned loopDepth) { + // Limit the loop depth ridiculousness. + if (loopDepth > 200) + loopDepth = 200; + + // The loop depth is used to roughly estimate the number of times the + // instruction is executed. Something like 10^d is simple, but will quickly + // overflow a float. This expression behaves like 10^d for small d, but is + // more tempered for large d. At d=200 we get 6.7e33 which leaves a bit of + // headroom before overflow. + float lc = powf(1 + (100.0f / (loopDepth+10)), (float)loopDepth); + + return (isDef + isUse) * lc; +} + +void +LiveIntervals::normalizeSpillWeights(std::vector<LiveInterval*> &NewLIs) { + for (unsigned i = 0, e = NewLIs.size(); i != e; ++i) + normalizeSpillWeight(*NewLIs[i]); +} + std::vector<LiveInterval*> LiveIntervals:: addIntervalsForSpillsFast(const LiveInterval &li, const MachineLoopInfo *loopInfo, @@ -1570,8 +1572,7 @@ addIntervalsForSpillsFast(const LiveInterval &li, std::vector<LiveInterval*> added; - assert(li.weight != HUGE_VALF && - "attempt to spill already spilled interval!"); + assert(li.isSpillable() && "attempt to spill already spilled interval!"); DEBUG({ dbgs() << "\t\t\t\tadding intervals for spills for interval: "; @@ -1607,10 +1608,7 @@ addIntervalsForSpillsFast(const LiveInterval &li, // create a new register for this spill LiveInterval &nI = getOrCreateInterval(NewVReg); - - // the spill weight is now infinity as it - // cannot be spilled again - nI.weight = HUGE_VALF; + nI.markNotSpillable(); // Rewrite register operands to use the new vreg. for (SmallVectorImpl<unsigned>::iterator I = Indices.begin(), @@ -1664,8 +1662,7 @@ addIntervalsForSpills(const LiveInterval &li, if (EnableFastSpilling) return addIntervalsForSpillsFast(li, loopInfo, vrm); - assert(li.weight != HUGE_VALF && - "attempt to spill already spilled interval!"); + assert(li.isSpillable() && "attempt to spill already spilled interval!"); DEBUG({ dbgs() << "\t\t\t\tadding intervals for spills for interval: "; @@ -1739,6 +1736,7 @@ addIntervalsForSpills(const LiveInterval &li, } handleSpilledImpDefs(li, vrm, rc, NewLIs); + normalizeSpillWeights(NewLIs); return NewLIs; } @@ -1814,6 +1812,7 @@ addIntervalsForSpills(const LiveInterval &li, // Insert spills / restores if we are splitting. if (!TrySplit) { handleSpilledImpDefs(li, vrm, rc, NewLIs); + normalizeSpillWeights(NewLIs); return NewLIs; } @@ -1930,11 +1929,10 @@ addIntervalsForSpills(const LiveInterval &li, unsigned ImpUse = getReMatImplicitUse(li, ReMatDefMI); if (ImpUse) { // Re-matting an instruction with virtual register use. Add the - // register as an implicit use on the use MI and update the register - // interval's spill weight to HUGE_VALF to prevent it from being - // spilled. + // register as an implicit use on the use MI and mark the register + // interval as unspillable. LiveInterval &ImpLi = getInterval(ImpUse); - ImpLi.weight = HUGE_VALF; + ImpLi.markNotSpillable(); MI->addOperand(MachineOperand::CreateReg(ImpUse, false, true)); } } @@ -1973,6 +1971,7 @@ addIntervalsForSpills(const LiveInterval &li, } handleSpilledImpDefs(li, vrm, rc, RetNewLIs); + normalizeSpillWeights(RetNewLIs); return RetNewLIs; } diff --git a/lib/CodeGen/LiveVariables.cpp b/lib/CodeGen/LiveVariables.cpp index 8a124dc..68c8539 100644 --- a/lib/CodeGen/LiveVariables.cpp +++ b/lib/CodeGen/LiveVariables.cpp @@ -510,6 +510,7 @@ bool LiveVariables::runOnMachineFunction(MachineFunction &mf) { PHIVarInfo = new SmallVector<unsigned, 4>[MF->getNumBlockIDs()]; std::fill(PhysRegDef, PhysRegDef + NumRegs, (MachineInstr*)0); std::fill(PhysRegUse, PhysRegUse + NumRegs, (MachineInstr*)0); + PHIJoins.clear(); /// Get some space for a respectable number of registers. VirtRegInfo.resize(64); diff --git a/lib/CodeGen/MachineBasicBlock.cpp b/lib/CodeGen/MachineBasicBlock.cpp index 655a0bf..64134ce 100644 --- a/lib/CodeGen/MachineBasicBlock.cpp +++ b/lib/CodeGen/MachineBasicBlock.cpp @@ -143,36 +143,6 @@ MachineBasicBlock::iterator MachineBasicBlock::getFirstTerminator() { return I; } -/// isOnlyReachableViaFallthough - Return true if this basic block has -/// exactly one predecessor and the control transfer mechanism between -/// the predecessor and this block is a fall-through. -bool MachineBasicBlock::isOnlyReachableByFallthrough() const { - // If this is a landing pad, it isn't a fall through. If it has no preds, - // then nothing falls through to it. - if (isLandingPad() || pred_empty()) - return false; - - // If there isn't exactly one predecessor, it can't be a fall through. - const_pred_iterator PI = pred_begin(), PI2 = PI; - ++PI2; - if (PI2 != pred_end()) - return false; - - // The predecessor has to be immediately before this block. - const MachineBasicBlock *Pred = *PI; - - if (!Pred->isLayoutSuccessor(this)) - return false; - - // If the block is completely empty, then it definitely does fall through. - if (Pred->empty()) - return true; - - // Otherwise, check the last instruction. - const MachineInstr &LastInst = Pred->back(); - return !LastInst.getDesc().isBarrier(); -} - void MachineBasicBlock::dump() const { print(dbgs()); } diff --git a/lib/CodeGen/MachineCSE.cpp b/lib/CodeGen/MachineCSE.cpp new file mode 100644 index 0000000..023ace2 --- /dev/null +++ b/lib/CodeGen/MachineCSE.cpp @@ -0,0 +1,208 @@ +//===-- MachineCSE.cpp - Machine Common Subexpression Elimination Pass ----===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This pass performs global common subexpression elimination on machine +// instructions using a scoped hash table based value numbering scheme. It +// must be run while the machine function is still in SSA form. +// +//===----------------------------------------------------------------------===// + +#define DEBUG_TYPE "machine-cse" +#include "llvm/CodeGen/Passes.h" +#include "llvm/CodeGen/MachineDominators.h" +#include "llvm/CodeGen/MachineInstr.h" +#include "llvm/CodeGen/MachineRegisterInfo.h" +#include "llvm/Target/TargetInstrInfo.h" +#include "llvm/ADT/ScopedHashTable.h" +#include "llvm/ADT/Statistic.h" +#include "llvm/Support/Debug.h" + +using namespace llvm; + +namespace llvm { + template<> struct DenseMapInfo<MachineInstr*> { + static inline MachineInstr *getEmptyKey() { + return 0; + } + + static inline MachineInstr *getTombstoneKey() { + return reinterpret_cast<MachineInstr*>(-1); + } + + static unsigned getHashValue(const MachineInstr* const &MI) { + unsigned Hash = MI->getOpcode() * 37; + for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) { + const MachineOperand &MO = MI->getOperand(i); + uint64_t Key = (uint64_t)MO.getType() << 32; + switch (MO.getType()) { + default: break; + case MachineOperand::MO_Register: + if (MO.isDef() && TargetRegisterInfo::isVirtualRegister(MO.getReg())) + continue; // Skip virtual register defs. + Key |= MO.getReg(); + break; + case MachineOperand::MO_Immediate: + Key |= MO.getImm(); + break; + case MachineOperand::MO_FrameIndex: + case MachineOperand::MO_ConstantPoolIndex: + case MachineOperand::MO_JumpTableIndex: + Key |= MO.getIndex(); + break; + case MachineOperand::MO_MachineBasicBlock: + Key |= DenseMapInfo<void*>::getHashValue(MO.getMBB()); + break; + case MachineOperand::MO_GlobalAddress: + Key |= DenseMapInfo<void*>::getHashValue(MO.getGlobal()); + break; + case MachineOperand::MO_BlockAddress: + Key |= DenseMapInfo<void*>::getHashValue(MO.getBlockAddress()); + break; + } + Key += ~(Key << 32); + Key ^= (Key >> 22); + Key += ~(Key << 13); + Key ^= (Key >> 8); + Key += (Key << 3); + Key ^= (Key >> 15); + Key += ~(Key << 27); + Key ^= (Key >> 31); + Hash = (unsigned)Key + Hash * 37; + } + return Hash; + } + + static bool isEqual(const MachineInstr* const &LHS, + const MachineInstr* const &RHS) { + if (RHS == getEmptyKey() || RHS == getTombstoneKey() || + LHS == getEmptyKey() || LHS == getTombstoneKey()) + return LHS == RHS; + return LHS->isIdenticalTo(RHS, MachineInstr::IgnoreVRegDefs); + } + }; +} // end llvm namespace + +namespace { + class MachineCSE : public MachineFunctionPass { + const TargetInstrInfo *TII; + MachineRegisterInfo *MRI; + MachineDominatorTree *DT; + ScopedHashTable<MachineInstr*, unsigned> VNT; + unsigned CurrVN; + public: + static char ID; // Pass identification + MachineCSE() : MachineFunctionPass(&ID), CurrVN(0) {} + + virtual bool runOnMachineFunction(MachineFunction &MF); + + virtual void getAnalysisUsage(AnalysisUsage &AU) const { + AU.setPreservesCFG(); + MachineFunctionPass::getAnalysisUsage(AU); + AU.addRequired<MachineDominatorTree>(); + AU.addPreserved<MachineDominatorTree>(); + } + + private: + bool PerformTrivialCoalescing(MachineInstr *MI, MachineBasicBlock *MBB); + bool ProcessBlock(MachineDomTreeNode *Node); + }; +} // end anonymous namespace + +char MachineCSE::ID = 0; +static RegisterPass<MachineCSE> +X("machine-cse", "Machine Common Subexpression Elimination"); + +FunctionPass *llvm::createMachineCSEPass() { return new MachineCSE(); } + +bool MachineCSE::PerformTrivialCoalescing(MachineInstr *MI, + MachineBasicBlock *MBB) { + bool Changed = false; + for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) { + MachineOperand &MO = MI->getOperand(i); + if (MO.isReg() && MO.isUse()) { + unsigned Reg = MO.getReg(); + if (!Reg || TargetRegisterInfo::isPhysicalRegister(Reg)) + continue; + MachineInstr *DefMI = MRI->getVRegDef(Reg); + if (DefMI->getParent() == MBB) { + unsigned SrcReg, DstReg, SrcSubIdx, DstSubIdx; + if (TII->isMoveInstr(*DefMI, SrcReg, DstReg, SrcSubIdx, DstSubIdx) && + TargetRegisterInfo::isVirtualRegister(SrcReg) && + !SrcSubIdx && !DstSubIdx) { + MO.setReg(SrcReg); + Changed = true; + } + } + } + } + + return Changed; +} + +static bool hasLivePhysRegDefUse(MachineInstr *MI) { + for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) { + MachineOperand &MO = MI->getOperand(i); + if (!MO.isReg()) + continue; + unsigned Reg = MO.getReg(); + if (!Reg) + continue; + if (TargetRegisterInfo::isPhysicalRegister(Reg) && + !(MO.isDef() && MO.isDead())) + return true; + } + return false; +} + +bool MachineCSE::ProcessBlock(MachineDomTreeNode *Node) { + bool Changed = false; + + ScopedHashTableScope<MachineInstr*, unsigned> VNTS(VNT); + MachineBasicBlock *MBB = Node->getBlock(); + for (MachineBasicBlock::iterator I = MBB->begin(), E = MBB->end(); I != E; + ++I) { + MachineInstr *MI = &*I; + bool SawStore = false; + if (!MI->isSafeToMove(TII, 0, SawStore)) + continue; + // Ignore copies or instructions that read / write physical registers + // (except for dead defs of physical registers). + unsigned SrcReg, DstReg, SrcSubIdx, DstSubIdx; + if (TII->isMoveInstr(*MI, SrcReg, DstReg, SrcSubIdx, DstSubIdx)) + continue; + if (hasLivePhysRegDefUse(MI)) + continue; + + bool FoundCSE = VNT.count(MI); + if (!FoundCSE) { + // Look for trivial copy coalescing opportunities. + if (PerformTrivialCoalescing(MI, MBB)) + FoundCSE = VNT.count(MI); + } + + if (FoundCSE) + DEBUG(dbgs() << "Found a common subexpression: " << *MI); + else + VNT.insert(MI, ++CurrVN); + } + + // Recursively call ProcessBlock with childred. + const std::vector<MachineDomTreeNode*> &Children = Node->getChildren(); + for (unsigned i = 0, e = Children.size(); i != e; ++i) + Changed |= ProcessBlock(Children[i]); + + return Changed; +} + +bool MachineCSE::runOnMachineFunction(MachineFunction &MF) { + TII = MF.getTarget().getInstrInfo(); + MRI = &MF.getRegInfo(); + DT = &getAnalysis<MachineDominatorTree>(); + return ProcessBlock(DT->getRootNode()); +} diff --git a/lib/CodeGen/MachineFunction.cpp b/lib/CodeGen/MachineFunction.cpp index f141c56..4377d5b 100644 --- a/lib/CodeGen/MachineFunction.cpp +++ b/lib/CodeGen/MachineFunction.cpp @@ -95,6 +95,9 @@ MachineFunction::MachineFunction(Function *F, const TargetMachine &TM, MFInfo = 0; FrameInfo = new (Allocator.Allocate<MachineFrameInfo>()) MachineFrameInfo(*TM.getFrameInfo()); + if (Fn->hasFnAttr(Attribute::StackAlignment)) + FrameInfo->setMaxAlignment(Attribute::getStackAlignmentFromAttrs( + Fn->getAttributes().getFnAttributes())); ConstantPool = new (Allocator.Allocate<MachineConstantPool>()) MachineConstantPool(TM.getTargetData()); Alignment = TM.getTargetLowering()->getFunctionAlignment(F); diff --git a/lib/CodeGen/MachineInstr.cpp b/lib/CodeGen/MachineInstr.cpp index b6d98e8..cba93f1 100644 --- a/lib/CodeGen/MachineInstr.cpp +++ b/lib/CodeGen/MachineInstr.cpp @@ -18,6 +18,7 @@ #include "llvm/Type.h" #include "llvm/Value.h" #include "llvm/Assembly/Writer.h" +#include "llvm/CodeGen/MachineConstantPool.h" #include "llvm/CodeGen/MachineFunction.h" #include "llvm/CodeGen/MachineMemOperand.h" #include "llvm/CodeGen/MachineRegisterInfo.h" @@ -701,6 +702,28 @@ void MachineInstr::addMemOperand(MachineFunction &MF, MemRefsEnd = NewMemRefsEnd; } +bool MachineInstr::isIdenticalTo(const MachineInstr *Other, + MICheckType Check) const { + if (Other->getOpcode() != getOpcode() || + Other->getNumOperands() != getNumOperands()) + return false; + for (unsigned i = 0, e = getNumOperands(); i != e; ++i) { + const MachineOperand &MO = getOperand(i); + const MachineOperand &OMO = Other->getOperand(i); + if (Check != CheckDefs && MO.isReg() && MO.isDef()) { + if (Check == IgnoreDefs) + continue; + // Check == IgnoreVRegDefs + if (TargetRegisterInfo::isPhysicalRegister(MO.getReg()) || + TargetRegisterInfo::isPhysicalRegister(OMO.getReg())) + if (MO.getReg() != OMO.getReg()) + return false; + } else if (!MO.isIdenticalTo(OMO)) + return false; + } + return true; +} + /// removeFromParent - This method unlinks 'this' from the containing basic /// block, and returns it, but does not delete it. MachineInstr *MachineInstr::removeFromParent() { @@ -959,8 +982,8 @@ void MachineInstr::copyPredicates(const MachineInstr *MI) { /// SawStore is set to true, it means that there is a store (or call) between /// the instruction's location and its intended destination. bool MachineInstr::isSafeToMove(const TargetInstrInfo *TII, - bool &SawStore, - AliasAnalysis *AA) const { + AliasAnalysis *AA, + bool &SawStore) const { // Ignore stuff that we obviously can't move. if (TID->mayStore() || TID->isCall()) { SawStore = true; @@ -985,11 +1008,11 @@ bool MachineInstr::isSafeToMove(const TargetInstrInfo *TII, /// isSafeToReMat - Return true if it's safe to rematerialize the specified /// instruction which defined the specified register instead of copying it. bool MachineInstr::isSafeToReMat(const TargetInstrInfo *TII, - unsigned DstReg, - AliasAnalysis *AA) const { + AliasAnalysis *AA, + unsigned DstReg) const { bool SawStore = false; if (!TII->isTriviallyReMaterializable(this, AA) || - !isSafeToMove(TII, SawStore, AA)) + !isSafeToMove(TII, AA, SawStore)) return false; for (unsigned i = 0, e = getNumOperands(); i != e; ++i) { const MachineOperand &MO = getOperand(i); diff --git a/lib/CodeGen/MachineLICM.cpp b/lib/CodeGen/MachineLICM.cpp index 92c84f3..0361694 100644 --- a/lib/CodeGen/MachineLICM.cpp +++ b/lib/CodeGen/MachineLICM.cpp @@ -252,32 +252,6 @@ bool MachineLICM::IsLoopInvariantInst(MachineInstr &I) { return false; } - DEBUG({ - dbgs() << "--- Checking if we can hoist " << I; - if (I.getDesc().getImplicitUses()) { - dbgs() << " * Instruction has implicit uses:\n"; - - const TargetRegisterInfo *TRI = TM->getRegisterInfo(); - for (const unsigned *ImpUses = I.getDesc().getImplicitUses(); - *ImpUses; ++ImpUses) - dbgs() << " -> " << TRI->getName(*ImpUses) << "\n"; - } - - if (I.getDesc().getImplicitDefs()) { - dbgs() << " * Instruction has implicit defines:\n"; - - const TargetRegisterInfo *TRI = TM->getRegisterInfo(); - for (const unsigned *ImpDefs = I.getDesc().getImplicitDefs(); - *ImpDefs; ++ImpDefs) - dbgs() << " -> " << TRI->getName(*ImpDefs) << "\n"; - } - }); - - if (I.getDesc().getImplicitDefs() || I.getDesc().getImplicitUses()) { - DEBUG(dbgs() << "Cannot hoist with implicit defines or uses\n"); - return false; - } - // The instruction is loop invariant if all of its operands are. for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i) { const MachineOperand &MO = I.getOperand(i); @@ -311,6 +285,10 @@ bool MachineLICM::IsLoopInvariantInst(MachineInstr &I) { } else if (!MO.isDead()) { // A def that isn't dead. We can't move it. return false; + } else if (CurLoop->getHeader()->isLiveIn(Reg)) { + // If the reg is live into the loop, we can't hoist an instruction + // which would clobber it. + return false; } } @@ -467,7 +445,7 @@ MachineLICM::LookForDuplicate(const MachineInstr *MI, std::vector<const MachineInstr*> &PrevMIs) { for (unsigned i = 0, e = PrevMIs.size(); i != e; ++i) { const MachineInstr *PrevMI = PrevMIs[i]; - if (TII->isIdentical(MI, PrevMI, RegInfo)) + if (TII->produceSameValue(MI, PrevMI)) return PrevMI; } return 0; @@ -480,9 +458,20 @@ bool MachineLICM::EliminateCSE(MachineInstr *MI, if (const MachineInstr *Dup = LookForDuplicate(MI, CI->second)) { DEBUG(dbgs() << "CSEing " << *MI << " with " << *Dup); + + // Replace virtual registers defined by MI by their counterparts defined + // by Dup. for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) { const MachineOperand &MO = MI->getOperand(i); - if (MO.isReg() && MO.isDef()) + + // Physical registers may not differ here. + assert((!MO.isReg() || MO.getReg() == 0 || + !TargetRegisterInfo::isPhysicalRegister(MO.getReg()) || + MO.getReg() == Dup->getOperand(i).getReg()) && + "Instructions with different phys regs are not identical!"); + + if (MO.isReg() && MO.isDef() && + !TargetRegisterInfo::isPhysicalRegister(MO.getReg())) RegInfo->replaceRegWith(MO.getReg(), Dup->getOperand(i).getReg()); } MI->eraseFromParent(); diff --git a/lib/CodeGen/MachineSink.cpp b/lib/CodeGen/MachineSink.cpp index c391576..9ba7d14 100644 --- a/lib/CodeGen/MachineSink.cpp +++ b/lib/CodeGen/MachineSink.cpp @@ -149,7 +149,7 @@ bool MachineSinking::ProcessBlock(MachineBasicBlock &MBB) { /// instruction out of its current block into a successor. bool MachineSinking::SinkInstruction(MachineInstr *MI, bool &SawStore) { // Check if it's safe to move the instruction. - if (!MI->isSafeToMove(TII, SawStore, AA)) + if (!MI->isSafeToMove(TII, AA, SawStore)) return false; // FIXME: This should include support for sinking instructions within the diff --git a/lib/CodeGen/PBQP/HeuristicSolver.h b/lib/CodeGen/PBQP/HeuristicSolver.h index c156264..bd18b52 100644 --- a/lib/CodeGen/PBQP/HeuristicSolver.h +++ b/lib/CodeGen/PBQP/HeuristicSolver.h @@ -229,7 +229,7 @@ namespace PBQP { } /// \brief Apply rule R1. - /// @param nItr Node iterator for node to apply R1 to. + /// @param xnItr Node iterator for node to apply R1 to. /// /// Node will be automatically pushed to the solver stack. void applyR1(Graph::NodeItr xnItr) { @@ -277,7 +277,7 @@ namespace PBQP { } /// \brief Apply rule R2. - /// @param nItr Node iterator for node to apply R2 to. + /// @param xnItr Node iterator for node to apply R2 to. /// /// Node will be automatically pushed to the solver stack. void applyR2(Graph::NodeItr xnItr) { diff --git a/lib/CodeGen/PHIElimination.cpp b/lib/CodeGen/PHIElimination.cpp index b740c68..bdfd448 100644 --- a/lib/CodeGen/PHIElimination.cpp +++ b/lib/CodeGen/PHIElimination.cpp @@ -55,8 +55,6 @@ void llvm::PHIElimination::getAnalysisUsage(AnalysisUsage &AU) const { bool llvm::PHIElimination::runOnMachineFunction(MachineFunction &Fn) { MRI = &Fn.getRegInfo(); - PHIDefs.clear(); - PHIKills.clear(); bool Changed = false; // Split critical edges to help the coalescer @@ -215,10 +213,6 @@ void llvm::PHIElimination::LowerAtomicPHINode( TII->copyRegToReg(MBB, AfterPHIsIt, DestReg, IncomingReg, RC, RC); } - // Record PHI def. - assert(!hasPHIDef(DestReg) && "Vreg has multiple phi-defs?"); - PHIDefs[DestReg] = &MBB; - // Update live variable information if there is any. LiveVariables *LV = getAnalysisIfAvailable<LiveVariables>(); if (LV) { @@ -229,6 +223,7 @@ void llvm::PHIElimination::LowerAtomicPHINode( // Increment use count of the newly created virtual register. VI.NumUses++; + LV->setPHIJoin(IncomingReg); // When we are reusing the incoming register, it may already have been // killed in this block. The old kill will also have been inserted at @@ -276,9 +271,6 @@ void llvm::PHIElimination::LowerAtomicPHINode( // path the PHI. MachineBasicBlock &opBlock = *MPhi->getOperand(i*2+2).getMBB(); - // Record the kill. - PHIKills[SrcReg].insert(&opBlock); - // If source is defined by an implicit def, there is no need to insert a // copy. MachineInstr *DefMI = MRI->getVRegDef(SrcReg); diff --git a/lib/CodeGen/PHIElimination.h b/lib/CodeGen/PHIElimination.h index f3ab9e2..ff4aa20 100644 --- a/lib/CodeGen/PHIElimination.h +++ b/lib/CodeGen/PHIElimination.h @@ -22,17 +22,8 @@ namespace llvm { /// Lower PHI instructions to copies. class PHIElimination : public MachineFunctionPass { MachineRegisterInfo *MRI; // Machine register information - private: - - typedef SmallSet<MachineBasicBlock*, 4> PHIKillList; - typedef DenseMap<unsigned, PHIKillList> PHIKillMap; - typedef DenseMap<unsigned, MachineBasicBlock*> PHIDefMap; public: - - typedef PHIKillList::iterator phi_kill_iterator; - typedef PHIKillList::const_iterator const_phi_kill_iterator; - static char ID; // Pass identification, replacement for typeid PHIElimination() : MachineFunctionPass(&ID) {} @@ -40,38 +31,6 @@ namespace llvm { virtual void getAnalysisUsage(AnalysisUsage &AU) const; - /// Return true if the given vreg was defined by a PHI intsr prior to - /// lowering. - bool hasPHIDef(unsigned vreg) const { - return PHIDefs.count(vreg); - } - - /// Returns the block in which the PHI instruction which defined the - /// given vreg used to reside. - MachineBasicBlock* getPHIDefBlock(unsigned vreg) { - PHIDefMap::iterator phiDefItr = PHIDefs.find(vreg); - assert(phiDefItr != PHIDefs.end() && "vreg has no phi-def."); - return phiDefItr->second; - } - - /// Returns true if the given vreg was killed by a PHI instr. - bool hasPHIKills(unsigned vreg) const { - return PHIKills.count(vreg); - } - - /// Returns an iterator over the BasicBlocks which contained PHI - /// kills of this register prior to lowering. - phi_kill_iterator phiKillsBegin(unsigned vreg) { - PHIKillMap::iterator phiKillItr = PHIKills.find(vreg); - assert(phiKillItr != PHIKills.end() && "vreg has no phi-kills."); - return phiKillItr->second.begin(); - } - phi_kill_iterator phiKillsEnd(unsigned vreg) { - PHIKillMap::iterator phiKillItr = PHIKills.find(vreg); - assert(phiKillItr != PHIKills.end() && "vreg has no phi-kills."); - return phiKillItr->second.end(); - } - private: /// EliminatePHINodes - Eliminate phi nodes by inserting copy instructions /// in predecessor basic blocks. @@ -139,8 +98,6 @@ namespace llvm { typedef DenseMap<BBVRegPair, unsigned> VRegPHIUse; VRegPHIUse VRegPHIUseCount; - PHIDefMap PHIDefs; - PHIKillMap PHIKills; // Defs of PHI sources which are implicit_def. SmallPtrSet<MachineInstr*, 4> ImpDefs; diff --git a/lib/CodeGen/Passes.cpp b/lib/CodeGen/Passes.cpp index f67eb79..5ea2941 100644 --- a/lib/CodeGen/Passes.cpp +++ b/lib/CodeGen/Passes.cpp @@ -34,7 +34,7 @@ static cl::opt<RegisterRegAlloc::FunctionPassCtor, false, RegisterPassParser<RegisterRegAlloc> > RegAlloc("regalloc", cl::init(&createLinearScanRegisterAllocator), - cl::desc("Register allocator to use: (default = linearscan)")); + cl::desc("Register allocator to use (default=linearscan)")); //===---------------------------------------------------------------------===// diff --git a/lib/CodeGen/PrologEpilogInserter.cpp b/lib/CodeGen/PrologEpilogInserter.cpp index 040259e..138e711 100644 --- a/lib/CodeGen/PrologEpilogInserter.cpp +++ b/lib/CodeGen/PrologEpilogInserter.cpp @@ -175,9 +175,10 @@ void PEI::calculateCallsInformation(MachineFunction &Fn) { MachineBasicBlock::iterator I = *i; // If call frames are not being included as part of the stack frame, and - // there is no dynamic allocation (therefore referencing frame slots off - // sp), leave the pseudo ops alone. We'll eliminate them later. - if (RegInfo->hasReservedCallFrame(Fn) || RegInfo->hasFP(Fn)) + // the target doesn't indicate otherwise, remove the call frame pseudos + // here. The sub/add sp instruction pairs are still inserted, but we don't + // need to track the SP adjustment for frame index elimination. + if (RegInfo->canSimplifyCallFramePseudos(Fn)) RegInfo->eliminateCallFramePseudoInstr(Fn, *I->getParent(), I); } } diff --git a/lib/CodeGen/RegAllocLinearScan.cpp b/lib/CodeGen/RegAllocLinearScan.cpp index 8e44a57..5c5a394 100644 --- a/lib/CodeGen/RegAllocLinearScan.cpp +++ b/lib/CodeGen/RegAllocLinearScan.cpp @@ -334,10 +334,6 @@ namespace { SmallVector<unsigned, 256> &inactiveCounts, bool SkipDGRegs); - /// assignVirt2StackSlot - assigns this virtual register to a - /// stack slot. returns the stack slot - int assignVirt2StackSlot(unsigned virtReg); - void ComputeRelatedRegClasses(); template <typename ItTy> diff --git a/lib/CodeGen/RegAllocPBQP.cpp b/lib/CodeGen/RegAllocPBQP.cpp index 2701faf..81cfd8f 100644 --- a/lib/CodeGen/RegAllocPBQP.cpp +++ b/lib/CodeGen/RegAllocPBQP.cpp @@ -57,7 +57,7 @@ using namespace llvm; static RegisterRegAlloc -registerPBQPRepAlloc("pbqp", "PBQP register allocator.", +registerPBQPRepAlloc("pbqp", "PBQP register allocator", llvm::createPBQPRegisterAllocator); static cl::opt<bool> @@ -867,10 +867,6 @@ bool PBQPRegAlloc::runOnMachineFunction(MachineFunction &MF) { // Find the vreg intervals in need of allocation. findVRegIntervalsToAlloc(); - // If there aren't any then we're done here. - if (vregIntervalsToAlloc.empty() && emptyVRegIntervals.empty()) - return true; - // If there are non-empty intervals allocate them using pbqp. if (!vregIntervalsToAlloc.empty()) { diff --git a/lib/CodeGen/ScheduleDAGInstrs.cpp b/lib/CodeGen/ScheduleDAGInstrs.cpp index 56dd533..badf34e 100644 --- a/lib/CodeGen/ScheduleDAGInstrs.cpp +++ b/lib/CodeGen/ScheduleDAGInstrs.cpp @@ -72,7 +72,7 @@ static const Value *getUnderlyingObjectFromInt(const Value *V) { } else { return V; } - assert(isa<IntegerType>(V->getType()) && "Unexpected operand type!"); + assert(V->getType()->isIntegerTy() && "Unexpected operand type!"); } while (1); } @@ -87,7 +87,7 @@ static const Value *getUnderlyingObject(const Value *V) { break; const Value *O = getUnderlyingObjectFromInt(cast<User>(V)->getOperand(0)); // If that succeeded in finding a pointer, continue the search. - if (!isa<PointerType>(O->getType())) + if (!O->getType()->isPointerTy()) break; V = O; } while (1); diff --git a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp index 7da7848..e4ff44d 100644 --- a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp +++ b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp @@ -1064,7 +1064,7 @@ SDValue DAGCombiner::visitADD(SDNode *N) { if (VT.isInteger() && !VT.isVector()) { APInt LHSZero, LHSOne; APInt RHSZero, RHSOne; - APInt Mask = APInt::getAllOnesValue(VT.getSizeInBits()); + APInt Mask = APInt::getAllOnesValue(VT.getScalarType().getSizeInBits()); DAG.ComputeMaskedBits(N0, Mask, LHSZero, LHSOne); if (LHSZero.getBoolValue()) { @@ -1136,7 +1136,7 @@ SDValue DAGCombiner::visitADDC(SDNode *N) { // fold (addc a, b) -> (or a, b), CARRY_FALSE iff a and b share no bits. APInt LHSZero, LHSOne; APInt RHSZero, RHSOne; - APInt Mask = APInt::getAllOnesValue(VT.getSizeInBits()); + APInt Mask = APInt::getAllOnesValue(VT.getScalarType().getSizeInBits()); DAG.ComputeMaskedBits(N0, Mask, LHSZero, LHSOne); if (LHSZero.getBoolValue()) { @@ -1786,7 +1786,7 @@ SDValue DAGCombiner::visitAND(SDNode *N) { SDValue RAND = ReassociateOps(ISD::AND, N->getDebugLoc(), N0, N1); if (RAND.getNode() != 0) return RAND; - // fold (and (or x, 0xFFFF), 0xFF) -> 0xFF + // fold (and (or x, C), D) -> D if (C & D) == D if (N1C && N0.getOpcode() == ISD::OR) if (ConstantSDNode *ORI = dyn_cast<ConstantSDNode>(N0.getOperand(1))) if ((ORI->getAPIntValue() & N1C->getAPIntValue()) == N1C->getAPIntValue()) @@ -2025,13 +2025,15 @@ SDValue DAGCombiner::visitOR(SDNode *N) { if (ROR.getNode() != 0) return ROR; // Canonicalize (or (and X, c1), c2) -> (and (or X, c2), c1|c2) + // iff (c1 & c2) == 0. if (N1C && N0.getOpcode() == ISD::AND && N0.getNode()->hasOneUse() && isa<ConstantSDNode>(N0.getOperand(1))) { ConstantSDNode *C1 = cast<ConstantSDNode>(N0.getOperand(1)); - return DAG.getNode(ISD::AND, N->getDebugLoc(), VT, - DAG.getNode(ISD::OR, N0.getDebugLoc(), VT, - N0.getOperand(0), N1), - DAG.FoldConstantArithmetic(ISD::OR, VT, N1C, C1)); + if ((C1->getAPIntValue() & N1C->getAPIntValue()) != 0) + return DAG.getNode(ISD::AND, N->getDebugLoc(), VT, + DAG.getNode(ISD::OR, N0.getDebugLoc(), VT, + N0.getOperand(0), N1), + DAG.FoldConstantArithmetic(ISD::OR, VT, N1C, C1)); } // fold (or (setcc x), (setcc y)) -> (setcc (or x, y)) if (isSetCCEquivalent(N0, LL, LR, CC0) && isSetCCEquivalent(N1, RL, RR, CC1)){ @@ -2754,7 +2756,7 @@ SDValue DAGCombiner::visitSRL(SDNode *N) { if (N1C && N0.getOpcode() == ISD::CTLZ && N1C->getAPIntValue() == Log2_32(VT.getSizeInBits())) { APInt KnownZero, KnownOne; - APInt Mask = APInt::getAllOnesValue(VT.getSizeInBits()); + APInt Mask = APInt::getAllOnesValue(VT.getScalarType().getSizeInBits()); DAG.ComputeMaskedBits(N0.getOperand(0), Mask, KnownZero, KnownOne); // If any of the input bits are KnownOne, then the input couldn't be all @@ -4655,7 +4657,8 @@ SDValue DAGCombiner::visitBRCOND(SDNode *N) { DAG.DeleteNode(Trunc); } // Replace the uses of SRL with SETCC - DAG.ReplaceAllUsesOfValueWith(N1, SetCC); + WorkListRemover DeadNodes(*this); + DAG.ReplaceAllUsesOfValueWith(N1, SetCC, &DeadNodes); removeFromWorkList(N1.getNode()); DAG.DeleteNode(N1.getNode()); return SDValue(N, 0); // Return N so it doesn't get rechecked! @@ -4663,6 +4666,56 @@ SDValue DAGCombiner::visitBRCOND(SDNode *N) { } } } + + // Transform br(xor(x, y)) -> br(x != y) + // Transform br(xor(xor(x,y), 1)) -> br (x == y) + if (N1.hasOneUse() && N1.getOpcode() == ISD::XOR) { + SDNode *TheXor = N1.getNode(); + SDValue Op0 = TheXor->getOperand(0); + SDValue Op1 = TheXor->getOperand(1); + if (Op0.getOpcode() == Op1.getOpcode()) { + // Avoid missing important xor optimizations. + SDValue Tmp = visitXOR(TheXor); + if (Tmp.getNode()) { + DEBUG(dbgs() << "\nReplacing.8 "; + TheXor->dump(&DAG); + dbgs() << "\nWith: "; + Tmp.getNode()->dump(&DAG); + dbgs() << '\n'); + WorkListRemover DeadNodes(*this); + DAG.ReplaceAllUsesOfValueWith(N1, Tmp, &DeadNodes); + removeFromWorkList(TheXor); + DAG.DeleteNode(TheXor); + return DAG.getNode(ISD::BRCOND, N->getDebugLoc(), + MVT::Other, Chain, Tmp, N2); + } + } + + if (Op0.getOpcode() != ISD::SETCC && Op1.getOpcode() != ISD::SETCC) { + bool Equal = false; + if (ConstantSDNode *RHSCI = dyn_cast<ConstantSDNode>(Op0)) + if (RHSCI->getAPIntValue() == 1 && Op0.hasOneUse() && + Op0.getOpcode() == ISD::XOR) { + TheXor = Op0.getNode(); + Equal = true; + } + + EVT SetCCVT = N1.getValueType(); + if (LegalTypes) + SetCCVT = TLI.getSetCCResultType(SetCCVT); + SDValue SetCC = DAG.getSetCC(TheXor->getDebugLoc(), + SetCCVT, + Op0, Op1, + Equal ? ISD::SETEQ : ISD::SETNE); + // Replace the uses of XOR with SETCC + WorkListRemover DeadNodes(*this); + DAG.ReplaceAllUsesOfValueWith(N1, SetCC, &DeadNodes); + removeFromWorkList(N1.getNode()); + DAG.DeleteNode(N1.getNode()); + return DAG.getNode(ISD::BRCOND, N->getDebugLoc(), + MVT::Other, Chain, SetCC, N2); + } + } return SDValue(); } @@ -5012,7 +5065,7 @@ SDValue DAGCombiner::visitLOAD(SDNode *N) { assert(N->getValueType(2) == MVT::Other && "Malformed indexed loads?"); if (N->hasNUsesOfValue(0, 0) && N->hasNUsesOfValue(0, 1)) { SDValue Undef = DAG.getUNDEF(N->getValueType(0)); - DEBUG(dbgs() << "\nReplacing.6 "; + DEBUG(dbgs() << "\nReplacing.7 "; N->dump(&DAG); dbgs() << "\nWith: "; Undef.getNode()->dump(&DAG); diff --git a/lib/CodeGen/SelectionDAG/FastISel.cpp b/lib/CodeGen/SelectionDAG/FastISel.cpp index 35ef5b7..1d76c7c 100644 --- a/lib/CodeGen/SelectionDAG/FastISel.cpp +++ b/lib/CodeGen/SelectionDAG/FastISel.cpp @@ -350,6 +350,34 @@ bool FastISel::SelectCall(User *I) { (void)TargetSelectInstruction(cast<Instruction>(I)); return true; } + case Intrinsic::dbg_value: { + // This requires target support, but right now X86 is the only Fast target. + DbgValueInst *DI = cast<DbgValueInst>(I); + const TargetInstrDesc &II = TII.get(TargetOpcode::DBG_VALUE); + Value *V = DI->getValue(); + if (!V) { + // Currently the optimizer can produce this; insert an undef to + // help debugging. Probably the optimizer should not do this. + BuildMI(MBB, DL, II).addReg(0U).addImm(DI->getOffset()). + addMetadata(DI->getVariable()); + } else if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) { + BuildMI(MBB, DL, II).addImm(CI->getZExtValue()).addImm(DI->getOffset()). + addMetadata(DI->getVariable()); + } else if (ConstantFP *CF = dyn_cast<ConstantFP>(V)) { + BuildMI(MBB, DL, II).addFPImm(CF).addImm(DI->getOffset()). + addMetadata(DI->getVariable()); + } else if (unsigned Reg = lookUpRegForValue(V)) { + BuildMI(MBB, DL, II).addReg(Reg, RegState::Debug).addImm(DI->getOffset()). + addMetadata(DI->getVariable()); + } else { + // We can't yet handle anything else here because it would require + // generating code, thus altering codegen because of debug info. + // Insert an undef so we can see what we dropped. + BuildMI(MBB, DL, II).addReg(0U).addImm(DI->getOffset()). + addMetadata(DI->getVariable()); + } + return true; + } case Intrinsic::eh_exception: { EVT VT = TLI.getValueType(I->getType()); switch (TLI.getOperationAction(ISD::EXCEPTIONADDR, VT)) { diff --git a/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp b/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp index e9321da..c7ab34f 100644 --- a/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp +++ b/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp @@ -1539,7 +1539,6 @@ SDValue SelectionDAGLegalize::ExpandVectorBuildThroughStack(SDNode* Node) { // the result as a vector. // Create the stack frame object. EVT VT = Node->getValueType(0); - EVT OpVT = Node->getOperand(0).getValueType(); EVT EltVT = VT.getVectorElementType(); DebugLoc dl = Node->getDebugLoc(); SDValue FIPtr = DAG.CreateStackTemporary(VT); @@ -1559,8 +1558,9 @@ SDValue SelectionDAGLegalize::ExpandVectorBuildThroughStack(SDNode* Node) { SDValue Idx = DAG.getConstant(Offset, FIPtr.getValueType()); Idx = DAG.getNode(ISD::ADD, dl, FIPtr.getValueType(), FIPtr, Idx); - // If EltVT smaller than OpVT, only store the bits necessary. - if (!OpVT.isVector() && EltVT.bitsLT(OpVT)) { + // If the destination vector element type is narrower than the source + // element type, only store the bits necessary. + if (EltVT.bitsLT(Node->getOperand(i).getValueType().getScalarType())) { Stores.push_back(DAG.getTruncStore(DAG.getEntryNode(), dl, Node->getOperand(i), Idx, SV, Offset, EltVT, false, false, 0)); @@ -1899,8 +1899,7 @@ SDValue SelectionDAGLegalize::ExpandLibCall(RTLIB::Libcall LC, SDNode *Node, TLI.LowerCallTo(InChain, RetTy, isSigned, !isSigned, false, false, 0, TLI.getLibcallCallingConv(LC), false, /*isReturnValueUsed=*/true, - Callee, Args, DAG, - Node->getDebugLoc(), DAG.GetOrdering(Node)); + Callee, Args, DAG, Node->getDebugLoc()); // Legalize the call sequence, starting with the chain. This will advance // the LastCALLSEQ_END to the legalized version of the CALLSEQ_END node that @@ -2308,7 +2307,7 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node, false, false, false, false, 0, CallingConv::C, false, /*isReturnValueUsed=*/true, DAG.getExternalSymbol("abort", TLI.getPointerTy()), - Args, DAG, dl, DAG.GetOrdering(Node)); + Args, DAG, dl); Results.push_back(CallResult.second); break; } diff --git a/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp b/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp index e4d123f..81f28ad 100644 --- a/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp +++ b/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp @@ -1080,8 +1080,8 @@ ExpandShiftWithKnownAmountBit(SDNode *N, SDValue &Lo, SDValue &Hi) { SDValue Amt = N->getOperand(1); EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0)); EVT ShTy = Amt.getValueType(); - unsigned ShBits = ShTy.getSizeInBits(); - unsigned NVTBits = NVT.getSizeInBits(); + unsigned ShBits = ShTy.getScalarType().getSizeInBits(); + unsigned NVTBits = NVT.getScalarType().getSizeInBits(); assert(isPowerOf2_32(NVTBits) && "Expanded integer type size not a power of two!"); DebugLoc dl = N->getDebugLoc(); diff --git a/lib/CodeGen/SelectionDAG/LegalizeTypes.cpp b/lib/CodeGen/SelectionDAG/LegalizeTypes.cpp index 0d929f1..f3e7ca4f 100644 --- a/lib/CodeGen/SelectionDAG/LegalizeTypes.cpp +++ b/lib/CodeGen/SelectionDAG/LegalizeTypes.cpp @@ -1034,8 +1034,7 @@ SDValue DAGTypeLegalizer::MakeLibCall(RTLIB::Libcall LC, EVT RetVT, TLI.LowerCallTo(DAG.getEntryNode(), RetTy, isSigned, !isSigned, false, false, 0, TLI.getLibcallCallingConv(LC), false, /*isReturnValueUsed=*/true, - Callee, Args, DAG, dl, - DAG.GetOrdering(DAG.getEntryNode().getNode())); + Callee, Args, DAG, dl); return CallInfo.first; } diff --git a/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp b/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp index b51c61b..06e7b8c 100644 --- a/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp +++ b/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp @@ -218,8 +218,20 @@ void ScheduleDAGSDNodes::BuildSchedUnits() { // Check to see if the scheduler cares about latencies. bool UnitLatencies = ForceUnitLatencies(); - for (SelectionDAG::allnodes_iterator NI = DAG->allnodes_begin(), - E = DAG->allnodes_end(); NI != E; ++NI) { + // Add all nodes in depth first order. + SmallVector<SDNode*, 64> Worklist; + SmallPtrSet<SDNode*, 64> Visited; + Worklist.push_back(DAG->getRoot().getNode()); + Visited.insert(DAG->getRoot().getNode()); + + while (!Worklist.empty()) { + SDNode *NI = Worklist.pop_back_val(); + + // Add all operands to the worklist unless they've already been added. + for (unsigned i = 0, e = NI->getNumOperands(); i != e; ++i) + if (Visited.insert(NI->getOperand(i).getNode())) + Worklist.push_back(NI->getOperand(i).getNode()); + if (isPassiveNode(NI)) // Leaf node, e.g. a TargetImmediate. continue; diff --git a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp index 43cf37e..002bc68 100644 --- a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp +++ b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp @@ -468,18 +468,20 @@ static void AddNodeIDNode(FoldingSetNodeID &ID, const SDNode *N) { } /// encodeMemSDNodeFlags - Generic routine for computing a value for use in -/// the CSE map that carries volatility, indexing mode, and +/// the CSE map that carries volatility, temporalness, indexing mode, and /// extension/truncation information. /// static inline unsigned -encodeMemSDNodeFlags(int ConvType, ISD::MemIndexedMode AM, bool isVolatile) { +encodeMemSDNodeFlags(int ConvType, ISD::MemIndexedMode AM, bool isVolatile, + bool isNonTemporal) { assert((ConvType & 3) == ConvType && "ConvType may not require more than 2 bits!"); assert((AM & 7) == AM && "AM may not require more than 3 bits!"); return ConvType | (AM << 2) | - (isVolatile << 5); + (isVolatile << 5) | + (isNonTemporal << 6); } //===----------------------------------------------------------------------===// @@ -860,14 +862,14 @@ SDValue SelectionDAG::getZeroExtendInReg(SDValue Op, DebugLoc DL, EVT VT) { /// getNOT - Create a bitwise NOT operation as (XOR Val, -1). /// SDValue SelectionDAG::getNOT(DebugLoc DL, SDValue Val, EVT VT) { - EVT EltVT = VT.isVector() ? VT.getVectorElementType() : VT; + EVT EltVT = VT.getScalarType(); SDValue NegOne = getConstant(APInt::getAllOnesValue(EltVT.getSizeInBits()), VT); return getNode(ISD::XOR, DL, VT, Val, NegOne); } SDValue SelectionDAG::getConstant(uint64_t Val, EVT VT, bool isT) { - EVT EltVT = VT.isVector() ? VT.getVectorElementType() : VT; + EVT EltVT = VT.getScalarType(); assert((EltVT.getSizeInBits() >= 64 || (uint64_t)((int64_t)Val >> EltVT.getSizeInBits()) + 1 < 2) && "getConstant with a uint64_t value that doesn't fit in the type!"); @@ -881,7 +883,7 @@ SDValue SelectionDAG::getConstant(const APInt &Val, EVT VT, bool isT) { SDValue SelectionDAG::getConstant(const ConstantInt &Val, EVT VT, bool isT) { assert(VT.isInteger() && "Cannot create FP integer constant!"); - EVT EltVT = VT.isVector() ? VT.getVectorElementType() : VT; + EVT EltVT = VT.getScalarType(); assert(Val.getBitWidth() == EltVT.getSizeInBits() && "APInt size does not match type size!"); @@ -924,8 +926,7 @@ SDValue SelectionDAG::getConstantFP(const APFloat& V, EVT VT, bool isTarget) { SDValue SelectionDAG::getConstantFP(const ConstantFP& V, EVT VT, bool isTarget){ assert(VT.isFloatingPoint() && "Cannot create integer FP constant!"); - EVT EltVT = - VT.isVector() ? VT.getVectorElementType() : VT; + EVT EltVT = VT.getScalarType(); // Do the map lookup using the actual bit pattern for the floating point // value, so that we don't have problems with 0.0 comparing equal to -0.0, and @@ -959,8 +960,7 @@ SDValue SelectionDAG::getConstantFP(const ConstantFP& V, EVT VT, bool isTarget){ } SDValue SelectionDAG::getConstantFP(double Val, EVT VT, bool isTarget) { - EVT EltVT = - VT.isVector() ? VT.getVectorElementType() : VT; + EVT EltVT = VT.getScalarType(); if (EltVT==MVT::f32) return getConstantFP(APFloat((float)Val), VT, isTarget); else @@ -1345,7 +1345,7 @@ SDValue SelectionDAG::getBlockAddress(BlockAddress *BA, EVT VT, } SDValue SelectionDAG::getSrcValue(const Value *V) { - assert((!V || isa<PointerType>(V->getType())) && + assert((!V || V->getType()->isPointerTy()) && "SrcValue is not a pointer?"); FoldingSetNodeID ID; @@ -2233,6 +2233,29 @@ bool SelectionDAG::isKnownNeverNaN(SDValue Op) const { return false; } +bool SelectionDAG::isKnownNeverZero(SDValue Op) const { + // If the value is a constant, we can obviously see if it is a zero or not. + if (const ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(Op)) + return !C->isZero(); + + // TODO: Recognize more cases here. + + return false; +} + +bool SelectionDAG::isEqualTo(SDValue A, SDValue B) const { + // Check the obvious case. + if (A == B) return true; + + // For for negative and positive zero. + if (const ConstantFPSDNode *CA = dyn_cast<ConstantFPSDNode>(A)) + if (const ConstantFPSDNode *CB = dyn_cast<ConstantFPSDNode>(B)) + if (CA->isZero() && CB->isZero()) return true; + + // Otherwise they may not be equal. + return false; +} + bool SelectionDAG::isVerifiedDebugInfoDesc(SDValue Op) const { GlobalAddressSDNode *GA = dyn_cast<GlobalAddressSDNode>(Op); if (!GA) return false; @@ -3081,8 +3104,7 @@ SDValue SelectionDAG::getStackArgumentTokenFactor(SDValue Chain) { /// operand. static SDValue getMemsetValue(SDValue Value, EVT VT, SelectionDAG &DAG, DebugLoc dl) { - unsigned NumBits = VT.isVector() ? - VT.getVectorElementType().getSizeInBits() : VT.getSizeInBits(); + unsigned NumBits = VT.getScalarType().getSizeInBits(); if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Value)) { APInt Val = APInt(NumBits, C->getZExtValue() & 255); unsigned Shift = 8; @@ -3474,7 +3496,7 @@ SDValue SelectionDAG::getMemcpy(SDValue Chain, DebugLoc dl, SDValue Dst, /*isReturnValueUsed=*/false, getExternalSymbol(TLI.getLibcallName(RTLIB::MEMCPY), TLI.getPointerTy()), - Args, *this, dl, GetOrdering(Chain.getNode())); + Args, *this, dl); return CallResult.second; } @@ -3523,7 +3545,7 @@ SDValue SelectionDAG::getMemmove(SDValue Chain, DebugLoc dl, SDValue Dst, /*isReturnValueUsed=*/false, getExternalSymbol(TLI.getLibcallName(RTLIB::MEMMOVE), TLI.getPointerTy()), - Args, *this, dl, GetOrdering(Chain.getNode())); + Args, *this, dl); return CallResult.second; } @@ -3582,7 +3604,7 @@ SDValue SelectionDAG::getMemset(SDValue Chain, DebugLoc dl, SDValue Dst, /*isReturnValueUsed=*/false, getExternalSymbol(TLI.getLibcallName(RTLIB::MEMSET), TLI.getPointerTy()), - Args, *this, dl, GetOrdering(Chain.getNode())); + Args, *this, dl); return CallResult.second; } @@ -3845,7 +3867,8 @@ SelectionDAG::getLoad(ISD::MemIndexedMode AM, DebugLoc dl, FoldingSetNodeID ID; AddNodeIDNode(ID, ISD::LOAD, VTs, Ops, 3); ID.AddInteger(MemVT.getRawBits()); - ID.AddInteger(encodeMemSDNodeFlags(ExtType, AM, MMO->isVolatile())); + ID.AddInteger(encodeMemSDNodeFlags(ExtType, AM, MMO->isVolatile(), + MMO->isNonTemporal())); void *IP = 0; if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) { cast<LoadSDNode>(E)->refineAlignment(MMO); @@ -3926,7 +3949,8 @@ SDValue SelectionDAG::getStore(SDValue Chain, DebugLoc dl, SDValue Val, FoldingSetNodeID ID; AddNodeIDNode(ID, ISD::STORE, VTs, Ops, 4); ID.AddInteger(VT.getRawBits()); - ID.AddInteger(encodeMemSDNodeFlags(false, ISD::UNINDEXED, MMO->isVolatile())); + ID.AddInteger(encodeMemSDNodeFlags(false, ISD::UNINDEXED, MMO->isVolatile(), + MMO->isNonTemporal())); void *IP = 0; if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) { cast<StoreSDNode>(E)->refineAlignment(MMO); @@ -3989,7 +4013,8 @@ SDValue SelectionDAG::getTruncStore(SDValue Chain, DebugLoc dl, SDValue Val, FoldingSetNodeID ID; AddNodeIDNode(ID, ISD::STORE, VTs, Ops, 4); ID.AddInteger(SVT.getRawBits()); - ID.AddInteger(encodeMemSDNodeFlags(true, ISD::UNINDEXED, MMO->isVolatile())); + ID.AddInteger(encodeMemSDNodeFlags(true, ISD::UNINDEXED, MMO->isVolatile(), + MMO->isNonTemporal())); void *IP = 0; if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) { cast<StoreSDNode>(E)->refineAlignment(MMO); @@ -4548,91 +4573,13 @@ SDNode *SelectionDAG::SelectNodeTo(SDNode *N, unsigned MachineOpc, SDNode *SelectionDAG::SelectNodeTo(SDNode *N, unsigned MachineOpc, SDVTList VTs, const SDValue *Ops, unsigned NumOps) { - return MorphNodeTo(N, ~MachineOpc, VTs, Ops, NumOps); -} - -SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc, - EVT VT) { - SDVTList VTs = getVTList(VT); - return MorphNodeTo(N, Opc, VTs, 0, 0); -} - -SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc, - EVT VT, SDValue Op1) { - SDVTList VTs = getVTList(VT); - SDValue Ops[] = { Op1 }; - return MorphNodeTo(N, Opc, VTs, Ops, 1); -} - -SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc, - EVT VT, SDValue Op1, - SDValue Op2) { - SDVTList VTs = getVTList(VT); - SDValue Ops[] = { Op1, Op2 }; - return MorphNodeTo(N, Opc, VTs, Ops, 2); -} - -SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc, - EVT VT, SDValue Op1, - SDValue Op2, SDValue Op3) { - SDVTList VTs = getVTList(VT); - SDValue Ops[] = { Op1, Op2, Op3 }; - return MorphNodeTo(N, Opc, VTs, Ops, 3); -} - -SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc, - EVT VT, const SDValue *Ops, - unsigned NumOps) { - SDVTList VTs = getVTList(VT); - return MorphNodeTo(N, Opc, VTs, Ops, NumOps); -} - -SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc, - EVT VT1, EVT VT2, const SDValue *Ops, - unsigned NumOps) { - SDVTList VTs = getVTList(VT1, VT2); - return MorphNodeTo(N, Opc, VTs, Ops, NumOps); -} - -SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc, - EVT VT1, EVT VT2) { - SDVTList VTs = getVTList(VT1, VT2); - return MorphNodeTo(N, Opc, VTs, (SDValue *)0, 0); -} - -SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc, - EVT VT1, EVT VT2, EVT VT3, - const SDValue *Ops, unsigned NumOps) { - SDVTList VTs = getVTList(VT1, VT2, VT3); - return MorphNodeTo(N, Opc, VTs, Ops, NumOps); -} - -SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc, - EVT VT1, EVT VT2, - SDValue Op1) { - SDVTList VTs = getVTList(VT1, VT2); - SDValue Ops[] = { Op1 }; - return MorphNodeTo(N, Opc, VTs, Ops, 1); -} - -SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc, - EVT VT1, EVT VT2, - SDValue Op1, SDValue Op2) { - SDVTList VTs = getVTList(VT1, VT2); - SDValue Ops[] = { Op1, Op2 }; - return MorphNodeTo(N, Opc, VTs, Ops, 2); -} - -SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc, - EVT VT1, EVT VT2, - SDValue Op1, SDValue Op2, - SDValue Op3) { - SDVTList VTs = getVTList(VT1, VT2); - SDValue Ops[] = { Op1, Op2, Op3 }; - return MorphNodeTo(N, Opc, VTs, Ops, 3); + N = MorphNodeTo(N, ~MachineOpc, VTs, Ops, NumOps); + // Reset the NodeID to -1. + N->setNodeId(-1); + return N; } -/// MorphNodeTo - These *mutate* the specified node to have the specified +/// MorphNodeTo - This *mutates* the specified node to have the specified /// return type, opcode, and operands. /// /// Note that MorphNodeTo returns the resultant node. If there is already a @@ -4708,12 +4655,14 @@ SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc, // Delete any nodes that are still dead after adding the uses for the // new operands. - SmallVector<SDNode *, 16> DeadNodes; - for (SmallPtrSet<SDNode *, 16>::iterator I = DeadNodeSet.begin(), - E = DeadNodeSet.end(); I != E; ++I) - if ((*I)->use_empty()) - DeadNodes.push_back(*I); - RemoveDeadNodes(DeadNodes); + if (!DeadNodeSet.empty()) { + SmallVector<SDNode *, 16> DeadNodes; + for (SmallPtrSet<SDNode *, 16>::iterator I = DeadNodeSet.begin(), + E = DeadNodeSet.end(); I != E; ++I) + if ((*I)->use_empty()) + DeadNodes.push_back(*I); + RemoveDeadNodes(DeadNodes); + } if (IP) CSEMap.InsertNode(N, IP); // Memoize the new node. @@ -5293,8 +5242,11 @@ GlobalAddressSDNode::GlobalAddressSDNode(unsigned Opc, const GlobalValue *GA, MemSDNode::MemSDNode(unsigned Opc, DebugLoc dl, SDVTList VTs, EVT memvt, MachineMemOperand *mmo) : SDNode(Opc, dl, VTs), MemoryVT(memvt), MMO(mmo) { - SubclassData = encodeMemSDNodeFlags(0, ISD::UNINDEXED, MMO->isVolatile()); + SubclassData = encodeMemSDNodeFlags(0, ISD::UNINDEXED, MMO->isVolatile(), + MMO->isNonTemporal()); assert(isVolatile() == MMO->isVolatile() && "Volatile encoding error!"); + assert(isNonTemporal() == MMO->isNonTemporal() && + "Non-temporal encoding error!"); assert(memvt.getStoreSize() == MMO->getSize() && "Size mismatch!"); } @@ -5303,7 +5255,8 @@ MemSDNode::MemSDNode(unsigned Opc, DebugLoc dl, SDVTList VTs, MachineMemOperand *mmo) : SDNode(Opc, dl, VTs, Ops, NumOps), MemoryVT(memvt), MMO(mmo) { - SubclassData = encodeMemSDNodeFlags(0, ISD::UNINDEXED, MMO->isVolatile()); + SubclassData = encodeMemSDNodeFlags(0, ISD::UNINDEXED, MMO->isVolatile(), + MMO->isNonTemporal()); assert(isVolatile() == MMO->isVolatile() && "Volatile encoding error!"); assert(memvt.getStoreSize() == MMO->getSize() && "Size mismatch!"); } @@ -5472,15 +5425,15 @@ std::string SDNode::getOperationName(const SelectionDAG *G) const { if (const TargetInstrInfo *TII = G->getTarget().getInstrInfo()) if (getMachineOpcode() < TII->getNumOpcodes()) return TII->get(getMachineOpcode()).getName(); - return "<<Unknown Machine Node>>"; + return "<<Unknown Machine Node #" + utostr(getOpcode()) + ">>"; } if (G) { const TargetLowering &TLI = G->getTargetLoweringInfo(); const char *Name = TLI.getTargetNodeName(getOpcode()); if (Name) return Name; - return "<<Unknown Target Node>>"; + return "<<Unknown Target Node #" + utostr(getOpcode()) + ">>"; } - return "<<Unknown Node>>"; + return "<<Unknown Node #" + utostr(getOpcode()) + ">>"; #ifndef NDEBUG case ISD::DELETED_NODE: @@ -5917,6 +5870,9 @@ void SDNode::print_details(raw_ostream &OS, const SelectionDAG *G) const { if (G) if (unsigned Order = G->GetOrdering(this)) OS << " [ORD=" << Order << ']'; + + if (getNodeId() != -1) + OS << " [ID=" << getNodeId() << ']'; } void SDNode::print(raw_ostream &OS, const SelectionDAG *G) const { @@ -6305,31 +6261,37 @@ bool ShuffleVectorSDNode::isSplatMask(const int *Mask, EVT VT) { return true; } +#ifdef XDEBUG static void checkForCyclesHelper(const SDNode *N, - std::set<const SDNode *> &visited) { - if (visited.find(N) != visited.end()) { + SmallPtrSet<const SDNode*, 32> &Visited, + SmallPtrSet<const SDNode*, 32> &Checked) { + // If this node has already been checked, don't check it again. + if (Checked.count(N)) + return; + + // If a node has already been visited on this depth-first walk, reject it as + // a cycle. + if (!Visited.insert(N)) { dbgs() << "Offending node:\n"; N->dumprFull(); - assert(0 && "Detected cycle in SelectionDAG"); + errs() << "Detected cycle in SelectionDAG\n"; + abort(); } - - std::set<const SDNode*>::iterator i; - bool inserted; - - tie(i, inserted) = visited.insert(N); - assert(inserted && "Missed cycle"); - - for(unsigned i = 0; i < N->getNumOperands(); ++i) { - checkForCyclesHelper(N->getOperand(i).getNode(), visited); - } - visited.erase(i); + + for(unsigned i = 0, e = N->getNumOperands(); i != e; ++i) + checkForCyclesHelper(N->getOperand(i).getNode(), Visited, Checked); + + Checked.insert(N); + Visited.erase(N); } +#endif void llvm::checkForCycles(const llvm::SDNode *N) { #ifdef XDEBUG assert(N && "Checking nonexistant SDNode"); - std::set<const SDNode *> visited; - checkForCyclesHelper(N, visited); + SmallPtrSet<const SDNode*, 32> visited; + SmallPtrSet<const SDNode*, 32> checked; + checkForCyclesHelper(N, visited, checked); #endif } diff --git a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp index 85ecb95..308742b 100644 --- a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp +++ b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp @@ -155,7 +155,7 @@ namespace { /// this value and returns the result as a ValueVTs value. This uses /// Chain/Flag as the input and updates them for the output Chain/Flag. /// If the Flag pointer is NULL, no flag is used. - SDValue getCopyFromRegs(SelectionDAG &DAG, DebugLoc dl, unsigned Order, + SDValue getCopyFromRegs(SelectionDAG &DAG, DebugLoc dl, SDValue &Chain, SDValue *Flag) const; /// getCopyToRegs - Emit a series of CopyToReg nodes that copies the @@ -163,14 +163,14 @@ namespace { /// Chain/Flag as the input and updates them for the output Chain/Flag. /// If the Flag pointer is NULL, no flag is used. void getCopyToRegs(SDValue Val, SelectionDAG &DAG, DebugLoc dl, - unsigned Order, SDValue &Chain, SDValue *Flag) const; + SDValue &Chain, SDValue *Flag) const; /// AddInlineAsmOperands - Add this value to the specified inlineasm node /// operand list. This adds the code marker, matching input operand index /// (if applicable), and includes the number of values added into it. void AddInlineAsmOperands(unsigned Code, bool HasMatching, unsigned MatchingIdx, - SelectionDAG &DAG, unsigned Order, + SelectionDAG &DAG, std::vector<SDValue> &Ops) const; }; } @@ -180,7 +180,7 @@ namespace { /// larger then ValueVT then AssertOp can be used to specify whether the extra /// bits are known to be zero (ISD::AssertZext) or sign extended from ValueVT /// (ISD::AssertSext). -static SDValue getCopyFromParts(SelectionDAG &DAG, DebugLoc dl, unsigned Order, +static SDValue getCopyFromParts(SelectionDAG &DAG, DebugLoc dl, const SDValue *Parts, unsigned NumParts, EVT PartVT, EVT ValueVT, ISD::NodeType AssertOp = ISD::DELETED_NODE) { @@ -205,9 +205,9 @@ static SDValue getCopyFromParts(SelectionDAG &DAG, DebugLoc dl, unsigned Order, EVT HalfVT = EVT::getIntegerVT(*DAG.getContext(), RoundBits/2); if (RoundParts > 2) { - Lo = getCopyFromParts(DAG, dl, Order, Parts, RoundParts / 2, + Lo = getCopyFromParts(DAG, dl, Parts, RoundParts / 2, PartVT, HalfVT); - Hi = getCopyFromParts(DAG, dl, Order, Parts + RoundParts / 2, + Hi = getCopyFromParts(DAG, dl, Parts + RoundParts / 2, RoundParts / 2, PartVT, HalfVT); } else { Lo = DAG.getNode(ISD::BIT_CONVERT, dl, HalfVT, Parts[0]); @@ -223,7 +223,7 @@ static SDValue getCopyFromParts(SelectionDAG &DAG, DebugLoc dl, unsigned Order, // Assemble the trailing non-power-of-2 part. unsigned OddParts = NumParts - RoundParts; EVT OddVT = EVT::getIntegerVT(*DAG.getContext(), OddParts * PartBits); - Hi = getCopyFromParts(DAG, dl, Order, + Hi = getCopyFromParts(DAG, dl, Parts + RoundParts, OddParts, PartVT, OddVT); // Combine the round and odd parts. @@ -259,7 +259,7 @@ static SDValue getCopyFromParts(SelectionDAG &DAG, DebugLoc dl, unsigned Order, // If the register was not expanded, truncate or copy the value, // as appropriate. for (unsigned i = 0; i != NumParts; ++i) - Ops[i] = getCopyFromParts(DAG, dl, Order, &Parts[i], 1, + Ops[i] = getCopyFromParts(DAG, dl, &Parts[i], 1, PartVT, IntermediateVT); } else if (NumParts > 0) { // If the intermediate type was expanded, build the intermediate @@ -268,7 +268,7 @@ static SDValue getCopyFromParts(SelectionDAG &DAG, DebugLoc dl, unsigned Order, "Must expand into a divisible number of parts!"); unsigned Factor = NumParts / NumIntermediates; for (unsigned i = 0; i != NumIntermediates; ++i) - Ops[i] = getCopyFromParts(DAG, dl, Order, &Parts[i * Factor], Factor, + Ops[i] = getCopyFromParts(DAG, dl, &Parts[i * Factor], Factor, PartVT, IntermediateVT); } @@ -292,7 +292,7 @@ static SDValue getCopyFromParts(SelectionDAG &DAG, DebugLoc dl, unsigned Order, assert(ValueVT.isFloatingPoint() && PartVT.isInteger() && !PartVT.isVector() && "Unexpected split"); EVT IntVT = EVT::getIntegerVT(*DAG.getContext(), ValueVT.getSizeInBits()); - Val = getCopyFromParts(DAG, dl, Order, Parts, NumParts, PartVT, IntVT); + Val = getCopyFromParts(DAG, dl, Parts, NumParts, PartVT, IntVT); } } @@ -349,7 +349,7 @@ static SDValue getCopyFromParts(SelectionDAG &DAG, DebugLoc dl, unsigned Order, /// getCopyToParts - Create a series of nodes that contain the specified value /// split into legal parts. If the parts contain more bits than Val, then, for /// integers, ExtendKind can be used to specify how to generate the extra bits. -static void getCopyToParts(SelectionDAG &DAG, DebugLoc dl, unsigned Order, +static void getCopyToParts(SelectionDAG &DAG, DebugLoc dl, SDValue Val, SDValue *Parts, unsigned NumParts, EVT PartVT, ISD::NodeType ExtendKind = ISD::ANY_EXTEND) { @@ -417,7 +417,7 @@ static void getCopyToParts(SelectionDAG &DAG, DebugLoc dl, unsigned Order, SDValue OddVal = DAG.getNode(ISD::SRL, dl, ValueVT, Val, DAG.getConstant(RoundBits, TLI.getPointerTy())); - getCopyToParts(DAG, dl, Order, OddVal, Parts + RoundParts, + getCopyToParts(DAG, dl, OddVal, Parts + RoundParts, OddParts, PartVT); if (TLI.isBigEndian()) @@ -514,7 +514,7 @@ static void getCopyToParts(SelectionDAG &DAG, DebugLoc dl, unsigned Order, // If the register was not expanded, promote or copy the value, // as appropriate. for (unsigned i = 0; i != NumParts; ++i) - getCopyToParts(DAG, dl, Order, Ops[i], &Parts[i], 1, PartVT); + getCopyToParts(DAG, dl, Ops[i], &Parts[i], 1, PartVT); } else if (NumParts > 0) { // If the intermediate type was expanded, split each the value into // legal parts. @@ -522,7 +522,7 @@ static void getCopyToParts(SelectionDAG &DAG, DebugLoc dl, unsigned Order, "Must expand into a divisible number of parts!"); unsigned Factor = NumParts / NumIntermediates; for (unsigned i = 0; i != NumIntermediates; ++i) - getCopyToParts(DAG, dl, Order, Ops[i], &Parts[i*Factor], Factor, PartVT); + getCopyToParts(DAG, dl, Ops[i], &Parts[i*Factor], Factor, PartVT); } } @@ -680,7 +680,7 @@ SDValue SelectionDAGBuilder::getValue(const Value *V) { getCurDebugLoc()); } - if (isa<StructType>(C->getType()) || isa<ArrayType>(C->getType())) { + if (C->getType()->isStructTy() || C->getType()->isArrayTy()) { assert((isa<ConstantAggregateZero>(C) || isa<UndefValue>(C)) && "Unknown struct or array constant!"); @@ -747,8 +747,7 @@ SDValue SelectionDAGBuilder::getValue(const Value *V) { RegsForValue RFV(*DAG.getContext(), TLI, InReg, V->getType()); SDValue Chain = DAG.getEntryNode(); - return RFV.getCopyFromRegs(DAG, getCurDebugLoc(), - SDNodeOrder, Chain, NULL); + return RFV.getCopyFromRegs(DAG, getCurDebugLoc(), Chain, NULL); } /// Get the EVTs and ArgFlags collections that represent the legalized return @@ -849,14 +848,12 @@ void SelectionDAGBuilder::visitRet(ReturnInst &I) { Chain = DAG.getNode(ISD::TokenFactor, getCurDebugLoc(), MVT::Other, &Chains[0], NumValues); - } else { - for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i) { - SmallVector<EVT, 4> ValueVTs; - ComputeValueVTs(TLI, I.getOperand(i)->getType(), ValueVTs); - unsigned NumValues = ValueVTs.size(); - if (NumValues == 0) continue; - - SDValue RetOp = getValue(I.getOperand(i)); + } else if (I.getNumOperands() != 0) { + SmallVector<EVT, 4> ValueVTs; + ComputeValueVTs(TLI, I.getOperand(0)->getType(), ValueVTs); + unsigned NumValues = ValueVTs.size(); + if (NumValues) { + SDValue RetOp = getValue(I.getOperand(0)); for (unsigned j = 0, f = NumValues; j != f; ++j) { EVT VT = ValueVTs[j]; @@ -881,7 +878,7 @@ void SelectionDAGBuilder::visitRet(ReturnInst &I) { unsigned NumParts = TLI.getNumRegisters(*DAG.getContext(), VT); EVT PartVT = TLI.getRegisterType(*DAG.getContext(), VT); SmallVector<SDValue, 4> Parts(NumParts); - getCopyToParts(DAG, getCurDebugLoc(), SDNodeOrder, + getCopyToParts(DAG, getCurDebugLoc(), SDValue(RetOp.getNode(), RetOp.getResNo() + j), &Parts[0], NumParts, PartVT, ExtendKind); @@ -2080,7 +2077,7 @@ void SelectionDAGBuilder::visitIndirectBr(IndirectBrInst &I) { void SelectionDAGBuilder::visitFSub(User &I) { // -0.0 - X --> fneg const Type *Ty = I.getType(); - if (isa<VectorType>(Ty)) { + if (Ty->isVectorTy()) { if (ConstantVector *CV = dyn_cast<ConstantVector>(I.getOperand(0))) { const VectorType *DestTy = cast<VectorType>(I.getType()); const Type *ElTy = DestTy->getElementType(); @@ -2117,7 +2114,7 @@ void SelectionDAGBuilder::visitBinary(User &I, unsigned OpCode) { void SelectionDAGBuilder::visitShift(User &I, unsigned Opcode) { SDValue Op1 = getValue(I.getOperand(0)); SDValue Op2 = getValue(I.getOperand(1)); - if (!isa<VectorType>(I.getType()) && + if (!I.getType()->isVectorTy() && Op2.getValueType() != TLI.getShiftAmountTy()) { // If the operand is smaller than the shift count type, promote it. EVT PTy = TLI.getPointerTy(); @@ -2890,7 +2887,7 @@ void SelectionDAGBuilder::visitTargetIntrinsic(CallInst &I, /// /// where Op is the hexidecimal representation of floating point value. static SDValue -GetSignificand(SelectionDAG &DAG, SDValue Op, DebugLoc dl, unsigned Order) { +GetSignificand(SelectionDAG &DAG, SDValue Op, DebugLoc dl) { SDValue t1 = DAG.getNode(ISD::AND, dl, MVT::i32, Op, DAG.getConstant(0x007fffff, MVT::i32)); SDValue t2 = DAG.getNode(ISD::OR, dl, MVT::i32, t1, @@ -2905,7 +2902,7 @@ GetSignificand(SelectionDAG &DAG, SDValue Op, DebugLoc dl, unsigned Order) { /// where Op is the hexidecimal representation of floating point value. static SDValue GetExponent(SelectionDAG &DAG, SDValue Op, const TargetLowering &TLI, - DebugLoc dl, unsigned Order) { + DebugLoc dl) { SDValue t0 = DAG.getNode(ISD::AND, dl, MVT::i32, Op, DAG.getConstant(0x7f800000, MVT::i32)); SDValue t1 = DAG.getNode(ISD::SRL, dl, MVT::i32, t0, @@ -3089,13 +3086,13 @@ SelectionDAGBuilder::visitLog(CallInst &I) { SDValue Op1 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, Op); // Scale the exponent by log(2) [0.69314718f]. - SDValue Exp = GetExponent(DAG, Op1, TLI, dl, SDNodeOrder); + SDValue Exp = GetExponent(DAG, Op1, TLI, dl); SDValue LogOfExponent = DAG.getNode(ISD::FMUL, dl, MVT::f32, Exp, getF32Constant(DAG, 0x3f317218)); // Get the significand and build it into a floating-point number with // exponent of 1. - SDValue X = GetSignificand(DAG, Op1, dl, SDNodeOrder); + SDValue X = GetSignificand(DAG, Op1, dl); if (LimitFloatPrecision <= 6) { // For floating-point precision of 6: @@ -3199,11 +3196,11 @@ SelectionDAGBuilder::visitLog2(CallInst &I) { SDValue Op1 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, Op); // Get the exponent. - SDValue LogOfExponent = GetExponent(DAG, Op1, TLI, dl, SDNodeOrder); + SDValue LogOfExponent = GetExponent(DAG, Op1, TLI, dl); // Get the significand and build it into a floating-point number with // exponent of 1. - SDValue X = GetSignificand(DAG, Op1, dl, SDNodeOrder); + SDValue X = GetSignificand(DAG, Op1, dl); // Different possible minimax approximations of significand in // floating-point for various degrees of accuracy over [1,2]. @@ -3308,13 +3305,13 @@ SelectionDAGBuilder::visitLog10(CallInst &I) { SDValue Op1 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, Op); // Scale the exponent by log10(2) [0.30102999f]. - SDValue Exp = GetExponent(DAG, Op1, TLI, dl, SDNodeOrder); + SDValue Exp = GetExponent(DAG, Op1, TLI, dl); SDValue LogOfExponent = DAG.getNode(ISD::FMUL, dl, MVT::f32, Exp, getF32Constant(DAG, 0x3e9a209a)); // Get the significand and build it into a floating-point number with // exponent of 1. - SDValue X = GetSignificand(DAG, Op1, dl, SDNodeOrder); + SDValue X = GetSignificand(DAG, Op1, dl); if (LimitFloatPrecision <= 6) { // For floating-point precision of 6: @@ -4287,8 +4284,8 @@ isInTailCallPosition(CallSite CS, Attributes CalleeRetAttr, // Check for a truly no-op bitcast. if (isa<BitCastInst>(U) && (U->getOperand(0)->getType() == U->getType() || - (isa<PointerType>(U->getOperand(0)->getType()) && - isa<PointerType>(U->getType())))) + (U->getOperand(0)->getType()->isPointerTy() && + U->getType()->isPointerTy()))) continue; // Otherwise it's not a true no-op. return false; @@ -4396,7 +4393,7 @@ void SelectionDAGBuilder::LowerCallTo(CallSite CS, SDValue Callee, CS.getCallingConv(), isTailCall, !CS.getInstruction()->use_empty(), - Callee, Args, DAG, getCurDebugLoc(), SDNodeOrder); + Callee, Args, DAG, getCurDebugLoc()); assert((isTailCall || Result.second.getNode()) && "Non-null chain expected with non-tail call!"); assert((Result.second.getNode() || !Result.first.getNode()) && @@ -4444,7 +4441,7 @@ void SelectionDAGBuilder::LowerCallTo(CallSite CS, SDValue Callee, unsigned NumRegs = TLI.getNumRegisters(RetTy->getContext(), VT); SDValue ReturnValue = - getCopyFromParts(DAG, getCurDebugLoc(), SDNodeOrder, &Values[CurReg], NumRegs, + getCopyFromParts(DAG, getCurDebugLoc(), &Values[CurReg], NumRegs, RegisterVT, VT, AssertOp); ReturnValues.push_back(ReturnValue); CurReg += NumRegs; @@ -4541,9 +4538,9 @@ bool SelectionDAGBuilder::visitMemCmpCall(CallInst &I) { return false; Value *LHS = I.getOperand(1), *RHS = I.getOperand(2); - if (!isa<PointerType>(LHS->getType()) || !isa<PointerType>(RHS->getType()) || - !isa<IntegerType>(I.getOperand(3)->getType()) || - !isa<IntegerType>(I.getType())) + if (!LHS->getType()->isPointerTy() || !RHS->getType()->isPointerTy() || + !I.getOperand(3)->getType()->isIntegerTy() || + !I.getType()->isIntegerTy()) return false; ConstantInt *Size = dyn_cast<ConstantInt>(I.getOperand(3)); @@ -4711,8 +4708,7 @@ void SelectionDAGBuilder::visitCall(CallInst &I) { /// Chain/Flag as the input and updates them for the output Chain/Flag. /// If the Flag pointer is NULL, no flag is used. SDValue RegsForValue::getCopyFromRegs(SelectionDAG &DAG, DebugLoc dl, - unsigned Order, SDValue &Chain, - SDValue *Flag) const { + SDValue &Chain, SDValue *Flag) const { // Assemble the legal parts into the final values. SmallVector<SDValue, 4> Values(ValueVTs.size()); SmallVector<SDValue, 8> Parts; @@ -4777,7 +4773,7 @@ SDValue RegsForValue::getCopyFromRegs(SelectionDAG &DAG, DebugLoc dl, Parts[i] = P; } - Values[Value] = getCopyFromParts(DAG, dl, Order, Parts.begin(), + Values[Value] = getCopyFromParts(DAG, dl, Parts.begin(), NumRegs, RegisterVT, ValueVT); Part += NumRegs; Parts.clear(); @@ -4793,8 +4789,7 @@ SDValue RegsForValue::getCopyFromRegs(SelectionDAG &DAG, DebugLoc dl, /// Chain/Flag as the input and updates them for the output Chain/Flag. /// If the Flag pointer is NULL, no flag is used. void RegsForValue::getCopyToRegs(SDValue Val, SelectionDAG &DAG, DebugLoc dl, - unsigned Order, SDValue &Chain, - SDValue *Flag) const { + SDValue &Chain, SDValue *Flag) const { // Get the list of the values's legal parts. unsigned NumRegs = Regs.size(); SmallVector<SDValue, 8> Parts(NumRegs); @@ -4803,7 +4798,7 @@ void RegsForValue::getCopyToRegs(SDValue Val, SelectionDAG &DAG, DebugLoc dl, unsigned NumParts = TLI->getNumRegisters(*DAG.getContext(), ValueVT); EVT RegisterVT = RegVTs[Value]; - getCopyToParts(DAG, dl, Order, + getCopyToParts(DAG, dl, Val.getValue(Val.getResNo() + Value), &Parts[Part], NumParts, RegisterVT); Part += NumParts; @@ -4844,7 +4839,7 @@ void RegsForValue::getCopyToRegs(SDValue Val, SelectionDAG &DAG, DebugLoc dl, /// values added into it. void RegsForValue::AddInlineAsmOperands(unsigned Code, bool HasMatching,unsigned MatchingIdx, - SelectionDAG &DAG, unsigned Order, + SelectionDAG &DAG, std::vector<SDValue> &Ops) const { assert(Regs.size() < (1 << 13) && "Too many inline asm outputs!"); unsigned Flag = Code | (Regs.size() << 3); @@ -5434,7 +5429,7 @@ void SelectionDAGBuilder::visitInlineAsm(CallSite CS) { 2 /* REGDEF */ , false, 0, - DAG, SDNodeOrder, + DAG, AsmNodeOperands); break; } @@ -5482,10 +5477,10 @@ void SelectionDAGBuilder::visitInlineAsm(CallSite CS) { // Use the produced MatchedRegs object to MatchedRegs.getCopyToRegs(InOperandVal, DAG, getCurDebugLoc(), - SDNodeOrder, Chain, &Flag); + Chain, &Flag); MatchedRegs.AddInlineAsmOperands(1 /*REGUSE*/, true, OpInfo.getMatchedOperand(), - DAG, SDNodeOrder, AsmNodeOperands); + DAG, AsmNodeOperands); break; } else { assert(((OpFlag & 7) == 4) && "Unknown matching constraint!"); @@ -5546,11 +5541,10 @@ void SelectionDAGBuilder::visitInlineAsm(CallSite CS) { } OpInfo.AssignedRegs.getCopyToRegs(InOperandVal, DAG, getCurDebugLoc(), - SDNodeOrder, Chain, &Flag); + Chain, &Flag); OpInfo.AssignedRegs.AddInlineAsmOperands(1/*REGUSE*/, false, 0, - DAG, SDNodeOrder, - AsmNodeOperands); + DAG, AsmNodeOperands); break; } case InlineAsm::isClobber: { @@ -5558,7 +5552,7 @@ void SelectionDAGBuilder::visitInlineAsm(CallSite CS) { // allocator is aware that the physreg got clobbered. if (!OpInfo.AssignedRegs.Regs.empty()) OpInfo.AssignedRegs.AddInlineAsmOperands(6 /* EARLYCLOBBER REGDEF */, - false, 0, DAG, SDNodeOrder, + false, 0, DAG, AsmNodeOperands); break; } @@ -5578,7 +5572,7 @@ void SelectionDAGBuilder::visitInlineAsm(CallSite CS) { // and set it as the value of the call. if (!RetValRegs.Regs.empty()) { SDValue Val = RetValRegs.getCopyFromRegs(DAG, getCurDebugLoc(), - SDNodeOrder, Chain, &Flag); + Chain, &Flag); // FIXME: Why don't we do this for inline asms with MRVs? if (CS.getType()->isSingleValueType() && CS.getType()->isSized()) { @@ -5618,7 +5612,7 @@ void SelectionDAGBuilder::visitInlineAsm(CallSite CS) { RegsForValue &OutRegs = IndirectStoresToEmit[i].first; Value *Ptr = IndirectStoresToEmit[i].second; SDValue OutVal = OutRegs.getCopyFromRegs(DAG, getCurDebugLoc(), - SDNodeOrder, Chain, &Flag); + Chain, &Flag); StoresToEmit.push_back(std::make_pair(OutVal, Ptr)); } @@ -5683,8 +5677,7 @@ TargetLowering::LowerCallTo(SDValue Chain, const Type *RetTy, CallingConv::ID CallConv, bool isTailCall, bool isReturnValueUsed, SDValue Callee, - ArgListTy &Args, SelectionDAG &DAG, DebugLoc dl, - unsigned Order) { + ArgListTy &Args, SelectionDAG &DAG, DebugLoc dl) { // Handle all of the outgoing arguments. SmallVector<ISD::OutputArg, 32> Outs; for (unsigned i = 0, e = Args.size(); i != e; ++i) { @@ -5735,7 +5728,7 @@ TargetLowering::LowerCallTo(SDValue Chain, const Type *RetTy, else if (Args[i].isZExt) ExtendKind = ISD::ZERO_EXTEND; - getCopyToParts(DAG, dl, Order, Op, &Parts[0], NumParts, + getCopyToParts(DAG, dl, Op, &Parts[0], NumParts, PartVT, ExtendKind); for (unsigned j = 0; j != NumParts; ++j) { @@ -5814,7 +5807,7 @@ TargetLowering::LowerCallTo(SDValue Chain, const Type *RetTy, EVT RegisterVT = getRegisterType(RetTy->getContext(), VT); unsigned NumRegs = getNumRegisters(RetTy->getContext(), VT); - ReturnValues.push_back(getCopyFromParts(DAG, dl, Order, &InVals[CurReg], + ReturnValues.push_back(getCopyFromParts(DAG, dl, &InVals[CurReg], NumRegs, RegisterVT, VT, AssertOp)); CurReg += NumRegs; @@ -5854,7 +5847,7 @@ void SelectionDAGBuilder::CopyValueToVirtualRegister(Value *V, unsigned Reg) { RegsForValue RFV(V->getContext(), TLI, Reg, V->getType()); SDValue Chain = DAG.getEntryNode(); - RFV.getCopyToRegs(Op, DAG, getCurDebugLoc(), SDNodeOrder, Chain, 0); + RFV.getCopyToRegs(Op, DAG, getCurDebugLoc(), Chain, 0); PendingExports.push_back(Chain); } @@ -5980,7 +5973,7 @@ void SelectionDAGISel::LowerArguments(BasicBlock *LLVMBB) { EVT VT = ValueVTs[0]; EVT RegVT = TLI.getRegisterType(*CurDAG->getContext(), VT); ISD::NodeType AssertOp = ISD::DELETED_NODE; - SDValue ArgValue = getCopyFromParts(DAG, dl, 0, &InVals[0], 1, + SDValue ArgValue = getCopyFromParts(DAG, dl, &InVals[0], 1, RegVT, VT, AssertOp); MachineFunction& MF = SDB->DAG.getMachineFunction(); @@ -6014,7 +6007,7 @@ void SelectionDAGISel::LowerArguments(BasicBlock *LLVMBB) { else if (F.paramHasAttr(Idx, Attribute::ZExt)) AssertOp = ISD::AssertZext; - ArgValues.push_back(getCopyFromParts(DAG, dl, 0, &InVals[i], + ArgValues.push_back(getCopyFromParts(DAG, dl, &InVals[i], NumParts, PartVT, VT, AssertOp)); } diff --git a/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp b/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp index eead526..2e2020d 100644 --- a/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp +++ b/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp @@ -723,9 +723,9 @@ void SelectionDAGISel::CodeGenAndEmitDAG() { // code to the MachineBasicBlock. if (TimePassesIsEnabled) { NamedRegionTimer T("Instruction Selection", GroupName); - InstructionSelect(); + DoInstructionSelection(); } else { - InstructionSelect(); + DoInstructionSelection(); } DEBUG(dbgs() << "Selected selection DAG:\n"); @@ -765,6 +765,66 @@ void SelectionDAGISel::CodeGenAndEmitDAG() { DEBUG(BB->dump()); } +void SelectionDAGISel::DoInstructionSelection() { + DEBUG(errs() << "===== Instruction selection begins:\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 + // node). + while (ISelPosition != CurDAG->allnodes_begin()) { + SDNode *Node = --ISelPosition; + // Skip dead nodes. DAGCombiner is expected to eliminate all dead nodes, + // but there are currently some corner cases that it misses. Also, this + // 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. + ISelUpdater ISU(ISelPosition); + CurDAG->RemoveDeadNode(Node, &ISU); + } + } + + CurDAG->setRoot(Dummy.getValue()); + } + DEBUG(errs() << "===== Instruction selection ends:\n"); + + PostprocessISelDAG(); + + // FIXME: This shouldn't be needed, remove it. + CurDAG->RemoveDeadNodes(); +} + + void SelectionDAGISel::SelectAllBasicBlocks(Function &Fn, MachineFunction &MF, MachineModuleInfo *MMI, @@ -1316,13 +1376,29 @@ static SDNode *findFlagUse(SDNode *N) { /// This function recursively traverses up the operand chain, ignoring /// certain nodes. static bool findNonImmUse(SDNode *Use, SDNode* Def, SDNode *ImmedUse, - SDNode *Root, - SmallPtrSet<SDNode*, 16> &Visited) { - if (Use->getNodeId() < Def->getNodeId() || - !Visited.insert(Use)) + SDNode *Root, SmallPtrSet<SDNode*, 16> &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 + // 'use' is smaller than the node we're scanning for, then we know we will + // 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 + // 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)) return false; for (unsigned i = 0, e = Use->getNumOperands(); i != e; ++i) { + // 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) @@ -1332,7 +1408,7 @@ static bool findNonImmUse(SDNode *Use, SDNode* Def, SDNode *ImmedUse, } // Traverse up the operand chain. - if (findNonImmUse(N, Def, ImmedUse, Root, Visited)) + if (findNonImmUse(N, Def, ImmedUse, Root, Visited, IgnoreChains)) return true; } return false; @@ -1347,9 +1423,10 @@ static bool findNonImmUse(SDNode *Use, SDNode* Def, SDNode *ImmedUse, /// have one non-chain use, we only need to watch out for load/op/store /// and load/op/cmp case where the root (store / cmp) may reach the load via /// its chain operand. -static inline bool isNonImmUse(SDNode *Root, SDNode *Def, SDNode *ImmedUse) { +static inline bool isNonImmUse(SDNode *Root, SDNode *Def, SDNode *ImmedUse, + bool IgnoreChains) { SmallPtrSet<SDNode*, 16> Visited; - return findNonImmUse(Root, Def, ImmedUse, Root, Visited); + return findNonImmUse(Root, Def, ImmedUse, Root, Visited, IgnoreChains); } /// IsProfitableToFold - Returns true if it's profitable to fold the specific @@ -1362,7 +1439,8 @@ bool SelectionDAGISel::IsProfitableToFold(SDValue N, SDNode *U, /// IsLegalToFold - Returns true if the specific operand node N of /// U can be folded during instruction selection that starts at Root. -bool SelectionDAGISel::IsLegalToFold(SDValue N, SDNode *U, SDNode *Root) const { +bool SelectionDAGISel::IsLegalToFold(SDValue N, SDNode *U, SDNode *Root, + bool IgnoreChains) const { if (OptLevel == CodeGenOpt::None) return false; // If Root use can somehow reach N through a path that that doesn't contain @@ -1416,7 +1494,7 @@ bool SelectionDAGISel::IsLegalToFold(SDValue N, SDNode *U, SDNode *Root) const { VT = Root->getValueType(Root->getNumValues()-1); } - return !isNonImmUse(Root, N.getNode(), U); + return !isNonImmUse(Root, N.getNode(), U, IgnoreChains); } SDNode *SelectionDAGISel::Select_INLINEASM(SDNode *N) { @@ -1428,6 +1506,7 @@ SDNode *SelectionDAGISel::Select_INLINEASM(SDNode *N) { VTs.push_back(MVT::Flag); SDValue New = CurDAG->getNode(ISD::INLINEASM, N->getDebugLoc(), VTs, &Ops[0], Ops.size()); + New->setNodeId(-1); return New.getNode(); } @@ -1443,7 +1522,1200 @@ SDNode *SelectionDAGISel::Select_EH_LABEL(SDNode *N) { MVT::Other, Tmp, Chain); } +/// GetVBR - decode a vbr encoding whose top bit is set. +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 { + NextBits = MatcherTable[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. +void SelectionDAGISel:: +UpdateChainsAndFlags(SDNode *NodeToMatch, SDValue InputChain, + const SmallVectorImpl<SDNode*> &ChainNodesMatched, + SDValue InputFlag, + const SmallVectorImpl<SDNode*> &FlagResultNodesMatched, + bool isMorphNodeTo) { + SmallVector<SDNode*, 4> NowDeadNodes; + + ISelUpdater ISU(ISelPosition); + + // Now that all the normal results are replaced, we replace the chain and + // flag results if present. + if (!ChainNodesMatched.empty()) { + assert(InputChain.getNode() != 0 && + "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. + 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) + ChainVal = ChainVal.getValue(ChainVal->getNumValues()-2); + assert(ChainVal.getValueType() == MVT::Other && "Not a chain?"); + CurDAG->ReplaceAllUsesOfValueWith(ChainVal, InputChain, &ISU); + + // If the node became dead, delete it. + if (ChainNode->use_empty()) + 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) { + // Handle any interior nodes explicitly marked. + for (unsigned i = 0, e = FlagResultNodesMatched.size(); i != e; ++i) { + SDNode *FRN = FlagResultNodesMatched[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"); + CurDAG->ReplaceAllUsesOfValueWith(SDValue(FRN, FRN->getNumValues()-1), + InputFlag, &ISU); + + // If the node became dead, delete it. + if (FRN->use_empty()) + NowDeadNodes.push_back(FRN); + } + } + + if (!NowDeadNodes.empty()) + CurDAG->RemoveDeadNodes(NowDeadNodes, &ISU); + + DEBUG(errs() << "ISEL: Match complete!\n"); +} + +enum ChainResult { + CR_Simple, + CR_InducesCycle, + CR_LeadsToInteriorNode +}; + +/// WalkChainUsers - Walk down the users of the specified chained node that is +/// part of the pattern we're matching, looking at all of the users we find. +/// This determines whether something is an interior node, whether we have a +/// non-pattern node in between two pattern nodes (which prevent folding because +/// it would induce a cycle) and whether we have a TokenFactor node sandwiched +/// between pattern nodes (in which case the TF becomes part of the pattern). +/// +/// 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, + SmallVectorImpl<SDNode*> &ChainedNodesInPattern, + SmallVectorImpl<SDNode*> &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 + // pattern that we're selecting down into the already selected chunk of the + // DAG. + 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) { + // If their node ID got reset to -1 then they've already been selected. + // Treat them like a MachineOpcode. + if (User->getNodeId() == -1) + continue; + } + + // If we have a TokenFactor, we handle it specially. + if (User->getOpcode() != ISD::TokenFactor) { + // If the node isn't a token factor and isn't part of our pattern, then it + // must be a random chained node in between two nodes we're selecting. + // This happens when we have something like: + // x = load ptr + // call + // y = x+4 + // store y -> ptr + // Because we structurally match the load/store as a read/modify/write, + // but the call is chained between them. We cannot fold in this case + // because it would induce a cycle in the graph. + 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 + // store y -> ptr + // This would happen when we're scanning down from the load and see the + // store as a user. Record that there is a use of ChainedNode that is + // part of the pattern and keep scanning uses. + Result = CR_LeadsToInteriorNode; + 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, + // the TokenFactor can be sandwiched in between two chained nodes, like so: + // [Load chain] + // ^ + // | + // [Load] + // ^ ^ + // | \ DAG's like cheese + // / \ do you? + // / | + // [TokenFactor] [Op] + // ^ ^ + // | | + // \ / + // \ / + // [Store] + // + // In this case, the TokenFactor becomes part of our match and we rewrite it + // as a new TokenFactor. + // + // To distinguish these two cases, do a recursive walk down the uses. + switch (WalkChainUsers(User, ChainedNodesInPattern, InteriorChainedNodes)) { + case CR_Simple: + // If the uses of the TokenFactor are just already-selected nodes, ignore + // it, it is "below" our pattern. + continue; + case CR_InducesCycle: + // If the uses of the TokenFactor lead to nodes that are not part of our + // pattern that are not selected, folding would turn this into a cycle, + // bail out now. + return CR_InducesCycle; + 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 + // ultimate chain result of the generated code. We will also add its chain + // inputs as inputs to the ultimate TokenFactor we create. + Result = CR_LeadsToInteriorNode; + ChainedNodesInPattern.push_back(User); + InteriorChainedNodes.push_back(User); + continue; + } + + return Result; +} + +/// HandleMergeInputChains - This implements the OPC_EmitMergeInputChains +/// operation for when the pattern matched at least one node with a chains. The +/// input vector contains a list of all of the chained nodes that we match. We +/// must determine if this is a valid thing to cover (i.e. matching it won't +/// induce cycles in the DAG) and if so, creating a TokenFactor node. that will +/// be used as the input node chain for the generated nodes. +static SDValue +HandleMergeInputChains(SmallVectorImpl<SDNode*> &ChainNodesMatched, + SelectionDAG *CurDAG) { + // Walk all of the chained nodes we've matched, recursively scanning down the + // users of the chain result. This adds any TokenFactor nodes that are caught + // in between chained nodes to the chained and interior nodes list. + SmallVector<SDNode*, 3> InteriorChainedNodes; + for (unsigned i = 0, e = ChainNodesMatched.size(); i != e; ++i) { + if (WalkChainUsers(ChainNodesMatched[i], 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<SDValue, 3> InputChains; + for (unsigned i = 0, e = ChainNodesMatched.size(); i != e; ++i) { + // Add the input chain of this node to the InputChains list (which will be + // the operands of the generated TokenFactor) if it's not an interior node. + SDNode *N = ChainNodesMatched[i]; + 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) { + if (!std::count(ChainNodesMatched.begin(), ChainNodesMatched.end(), + N->getOperand(op).getNode())) + 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:: +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. + // In this case we need to shifting the operands down. + // FIXME: This is a horrible hack and broken in obscure cases, no worse + // than the old isel though. We should sink this into MorphNodeTo. + int OldFlagResultNo = -1, OldChainResultNo = -1; + + unsigned NTMNumResults = Node->getNumValues(); + if (Node->getValueType(NTMNumResults-1) == MVT::Flag) { + OldFlagResultNo = NTMNumResults-1; + if (NTMNumResults != 1 && + Node->getValueType(NTMNumResults-2) == MVT::Other) + OldChainResultNo = NTMNumResults-2; + } else if (Node->getValueType(NTMNumResults-1) == MVT::Other) + OldChainResultNo = NTMNumResults-1; + + // 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); + + // MorphNodeTo can operate in two ways: if an existing node with the + // specified operands exists, it can just return it. Otherwise, it + // updates the node in place to have the requested operands. + if (Res == Node) { + // If we updated the node in place, reset the node ID. To the isel, + // this should be just like a newly allocated machine node. + Res->setNodeId(-1); + } + + unsigned ResNumResults = Res->getNumValues(); + // Move the flag if needed. + if ((EmitNodeInfo & OPFL_FlagOutput) && OldFlagResultNo != -1 && + (unsigned)OldFlagResultNo != ResNumResults-1) + CurDAG->ReplaceAllUsesOfValueWith(SDValue(Node, OldFlagResultNo), + SDValue(Res, ResNumResults-1)); + + if ((EmitNodeInfo & OPFL_FlagOutput) != 0) + --ResNumResults; + + // Move the chain reference if needed. + if ((EmitNodeInfo & OPFL_Chain) && OldChainResultNo != -1 && + (unsigned)OldChainResultNo != ResNumResults-1) + 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 +CheckSame(const unsigned char *MatcherTable, unsigned &MatcherIndex, + SDValue N, const SmallVectorImpl<SDValue> &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]; +} + +/// CheckPatternPredicate - Implements OP_CheckPatternPredicate. +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 +CheckNodePredicate(const unsigned char *MatcherTable, unsigned &MatcherIndex, + SelectionDAGISel &SDISel, SDNode *N) { + return SDISel.CheckNodePredicate(N, MatcherTable[MatcherIndex++]); +} + +ALWAYS_INLINE static bool +CheckOpcode(const unsigned char *MatcherTable, unsigned &MatcherIndex, + SDNode *N) { + return N->getOpcode() == MatcherTable[MatcherIndex++]; +} + +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 +CheckChildType(const unsigned char *MatcherTable, unsigned &MatcherIndex, + 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); +} + + +ALWAYS_INLINE static bool +CheckCondCode(const unsigned char *MatcherTable, unsigned &MatcherIndex, + SDValue N) { + return cast<CondCodeSDNode>(N)->get() == + (ISD::CondCode)MatcherTable[MatcherIndex++]; +} + +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<VTSDNode>(N)->getVT() == VT) + return true; + + // Handle the case when VT is iPTR. + return VT == MVT::iPTR && cast<VTSDNode>(N)->getVT() == TLI.getPointerTy(); +} + +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<ConstantSDNode>(N); + return C != 0 && C->getSExtValue() == Val; +} + +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<ConstantSDNode>(N->getOperand(1)); + return C != 0 && SDISel.CheckAndMask(N.getOperand(0), C, Val); +} + +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<ConstantSDNode>(N->getOperand(1)); + return C != 0 && SDISel.CheckOrMask(N.getOperand(0), C, Val); +} + +/// IsPredicateKnownToFail - If we know how and can do so without pushing a +/// scope, evaluate the current node. If the current predicate is known to +/// 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. +static unsigned IsPredicateKnownToFail(const unsigned char *Table, + unsigned Index, SDValue N, + bool &Result, SelectionDAGISel &SDISel, + SmallVectorImpl<SDValue> &RecordedNodes){ + switch (Table[Index++]) { + default: + Result = false; + return Index-1; // Could not evaluate this predicate. + case SelectionDAGISel::OPC_CheckSame: + Result = !::CheckSame(Table, Index, N, RecordedNodes); + return Index; + case SelectionDAGISel::OPC_CheckPatternPredicate: + Result = !::CheckPatternPredicate(Table, Index, SDISel); + return Index; + case SelectionDAGISel::OPC_CheckPredicate: + Result = !::CheckNodePredicate(Table, Index, SDISel, N.getNode()); + return Index; + case SelectionDAGISel::OPC_CheckOpcode: + Result = !::CheckOpcode(Table, Index, N.getNode()); + return Index; + case SelectionDAGISel::OPC_CheckType: + Result = !::CheckType(Table, Index, N, SDISel.TLI); + return Index; + case SelectionDAGISel::OPC_CheckChild0Type: + case SelectionDAGISel::OPC_CheckChild1Type: + case SelectionDAGISel::OPC_CheckChild2Type: + case SelectionDAGISel::OPC_CheckChild3Type: + case SelectionDAGISel::OPC_CheckChild4Type: + case SelectionDAGISel::OPC_CheckChild5Type: + case SelectionDAGISel::OPC_CheckChild6Type: + case SelectionDAGISel::OPC_CheckChild7Type: + Result = !::CheckChildType(Table, Index, N, SDISel.TLI, + Table[Index-1] - SelectionDAGISel::OPC_CheckChild0Type); + return Index; + case SelectionDAGISel::OPC_CheckCondCode: + Result = !::CheckCondCode(Table, Index, N); + return Index; + case SelectionDAGISel::OPC_CheckValueType: + Result = !::CheckValueType(Table, Index, N, SDISel.TLI); + return Index; + case SelectionDAGISel::OPC_CheckInteger: + Result = !::CheckInteger(Table, Index, N); + return Index; + case SelectionDAGISel::OPC_CheckAndImm: + Result = !::CheckAndImm(Table, Index, N, SDISel); + return Index; + case SelectionDAGISel::OPC_CheckOrImm: + Result = !::CheckOrImm(Table, Index, N, SDISel); + return Index; + } +} + + +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<SDValue, 4> 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; + + /// HasChainNodesMatched - True if the ChainNodesMatched list is non-empty. + bool HasChainNodesMatched, HasFlagResultNodesMatched; +}; + +SDNode *SelectionDAGISel:: +SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, + unsigned TableSize) { + // FIXME: Should these even be selected? Handle these cases in the caller? + switch (NodeToMatch->getOpcode()) { + default: + break; + case ISD::EntryToken: // These nodes remain the same. + case ISD::BasicBlock: + case ISD::Register: + case ISD::HANDLENODE: + case ISD::TargetConstant: + case ISD::TargetConstantFP: + case ISD::TargetConstantPool: + case ISD::TargetFrameIndex: + case ISD::TargetExternalSymbol: + case ISD::TargetBlockAddress: + case ISD::TargetJumpTable: + case ISD::TargetGlobalTLSAddress: + case ISD::TargetGlobalAddress: + case ISD::TokenFactor: + case ISD::CopyFromReg: + case ISD::CopyToReg: + NodeToMatch->setNodeId(-1); // Mark selected. + return 0; + case ISD::AssertSext: + case ISD::AssertZext: + CurDAG->ReplaceAllUsesOfValueWith(SDValue(NodeToMatch, 0), + NodeToMatch->getOperand(0)); + return 0; + case ISD::INLINEASM: return Select_INLINEASM(NodeToMatch); + case ISD::EH_LABEL: return Select_EH_LABEL(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. + SmallVector<SDValue, 8> NodeStack; + SDValue N = SDValue(NodeToMatch, 0); + NodeStack.push_back(N); + + // MatchScopes - Scopes used when matching, if a match failure happens, this + // indicates where to continue checking. + SmallVector<MatchScope, 8> MatchScopes; + + // RecordedNodes - This is the set of nodes that have been recorded by the + // state machine. + SmallVector<SDValue, 8> RecordedNodes; + + // MatchedMemRefs - This is the set of MemRef's we've seen in the input + // pattern. + SmallVector<MachineMemOperand*, 2> MatchedMemRefs; + + // These are the current input chain and flag for use when generating nodes. + // Various Emit operations change these. For example, emitting a copytoreg + // uses and updates these. + SDValue InputChain, InputFlag; + + // 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<SDNode*, 3> ChainNodesMatched; + SmallVector<SDNode*, 3> FlagResultNodesMatched; + + 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()) + MatcherIndex = OpcodeOffset[N.getOpcode()]; + DEBUG(errs() << " Initial Opcode index to " << MatcherIndex << "\n"); + + } else if (MatcherTable[0] == OPC_SwitchOpcode) { + // Otherwise, the table isn't computed, but the state machine does start + // with an OPC_SwitchOpcode instruction. Populate the table now, since this + // is the first time we're selecting an instruction. + unsigned Idx = 1; + while (1) { + // Get the size of this case. + unsigned CaseSize = MatcherTable[Idx++]; + if (CaseSize & 128) + CaseSize = GetVBR(CaseSize, MatcherTable, Idx); + if (CaseSize == 0) break; + + // Get the opcode, add the index to the table. + unsigned Opc = MatcherTable[Idx++]; + if (Opc >= OpcodeOffset.size()) + OpcodeOffset.resize((Opc+1)*2); + OpcodeOffset[Opc] = Idx; + Idx += CaseSize; + } + + // Okay, do the lookup for the first opcode. + if (N.getOpcode() < OpcodeOffset.size()) + MatcherIndex = OpcodeOffset[N.getOpcode()]; + } + + while (1) { + assert(MatcherIndex < TableSize && "Invalid index"); + BuiltinOpcodes Opcode = (BuiltinOpcodes)MatcherTable[MatcherIndex++]; + switch (Opcode) { + case OPC_Scope: { + // Okay, the semantics of this operation are that we should push a scope + // then evaluate the first child. However, pushing a scope only to have + // the first check fail (which then pops it) is inefficient. If we can + // 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) + NumToSkip = GetVBR(NumToSkip, MatcherTable, MatcherIndex); + // Found the end of the scope with no match. + if (NumToSkip == 0) { + FailIndex = 0; + break; + } + + FailIndex = MatcherIndex+NumToSkip; + + // 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. + bool Result; + MatcherIndex = IsPredicateKnownToFail(MatcherTable, MatcherIndex, N, + Result, *this, RecordedNodes); + if (!Result) + break; + + DEBUG(errs() << " Skipped scope entry at index " << MatcherIndex + << " continuing at " << FailIndex << "\n"); + + + // 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; + NewEntry.FailIndex = FailIndex; + NewEntry.NodeStack.append(NodeStack.begin(), NodeStack.end()); + NewEntry.NumRecordedNodes = RecordedNodes.size(); + NewEntry.NumMatchedMemRefs = MatchedMemRefs.size(); + NewEntry.InputChain = InputChain; + NewEntry.InputFlag = InputFlag; + NewEntry.HasChainNodesMatched = !ChainNodesMatched.empty(); + NewEntry.HasFlagResultNodesMatched = !FlagResultNodesMatched.empty(); + MatchScopes.push_back(NewEntry); + continue; + } + case OPC_RecordNode: + // Remember this node, it may end up being an operand in the pattern. + RecordedNodes.push_back(N); + continue; + + case OPC_RecordChild0: case OPC_RecordChild1: + case OPC_RecordChild2: case OPC_RecordChild3: + case OPC_RecordChild4: case OPC_RecordChild5: + case OPC_RecordChild6: case OPC_RecordChild7: { + unsigned ChildNo = Opcode-OPC_RecordChild0; + if (ChildNo >= N.getNumOperands()) + break; // Match fails if out of range child #. + + RecordedNodes.push_back(N->getOperand(ChildNo)); + continue; + } + case OPC_RecordMemRef: + MatchedMemRefs.push_back(cast<MemSDNode>(N)->getMemOperand()); + continue; + + case OPC_CaptureFlagInput: + // If the current node has an input flag, capture it in InputFlag. + if (N->getNumOperands() != 0 && + N->getOperand(N->getNumOperands()-1).getValueType() == MVT::Flag) + InputFlag = N->getOperand(N->getNumOperands()-1); + continue; + + case OPC_MoveChild: { + unsigned ChildNo = MatcherTable[MatcherIndex++]; + if (ChildNo >= N.getNumOperands()) + break; // Match fails if out of range child #. + N = N.getOperand(ChildNo); + 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(); + continue; + + case OPC_CheckSame: + if (!::CheckSame(MatcherTable, MatcherIndex, N, RecordedNodes)) break; + continue; + case OPC_CheckPatternPredicate: + if (!::CheckPatternPredicate(MatcherTable, MatcherIndex, *this)) break; + continue; + case OPC_CheckPredicate: + if (!::CheckNodePredicate(MatcherTable, MatcherIndex, *this, + N.getNode())) + break; + continue; + case OPC_CheckComplexPat: + if (!CheckComplexPattern(NodeToMatch, N, + MatcherTable[MatcherIndex++], RecordedNodes)) + break; + continue; + 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; + unsigned CaseSize; + while (1) { + // Get the size of this case. + CaseSize = MatcherTable[MatcherIndex++]; + if (CaseSize & 128) + CaseSize = GetVBR(CaseSize, MatcherTable, MatcherIndex); + if (CaseSize == 0) break; + + // If the opcode matches, then we will execute this case. + if (CurNodeOpcode == MatcherTable[MatcherIndex++]) + 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; + unsigned SwitchStart = MatcherIndex-1; (void)SwitchStart; + unsigned CaseSize; + while (1) { + // Get the size of this case. + CaseSize = MatcherTable[MatcherIndex++]; + if (CaseSize & 128) + CaseSize = GetVBR(CaseSize, MatcherTable, MatcherIndex); + if (CaseSize == 0) break; + + MVT::SimpleValueType CaseVT = + (MVT::SimpleValueType)MatcherTable[MatcherIndex++]; + if (CaseVT == MVT::iPTR) + CaseVT = TLI.getPointerTy().SimpleTy; + + // 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'); + continue; + } + case OPC_CheckChild0Type: case OPC_CheckChild1Type: + case OPC_CheckChild2Type: case OPC_CheckChild3Type: + case OPC_CheckChild4Type: case OPC_CheckChild5Type: + case OPC_CheckChild6Type: case OPC_CheckChild7Type: + if (!::CheckChildType(MatcherTable, MatcherIndex, N, TLI, + Opcode-OPC_CheckChild0Type)) + break; + continue; + case OPC_CheckCondCode: + if (!::CheckCondCode(MatcherTable, MatcherIndex, N)) break; + continue; + case OPC_CheckValueType: + if (!::CheckValueType(MatcherTable, MatcherIndex, N, TLI)) break; + continue; + case OPC_CheckInteger: + if (!::CheckInteger(MatcherTable, MatcherIndex, N)) break; + continue; + case OPC_CheckAndImm: + if (!::CheckAndImm(MatcherTable, MatcherIndex, N, *this)) break; + continue; + 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 + // a single use. + bool HasMultipleUses = false; + for (unsigned i = 1, e = NodeStack.size()-1; i != e; ++i) + if (!NodeStack[i].hasOneUse()) { + HasMultipleUses = true; + break; + } + if (HasMultipleUses) break; + + // Check to see that the target thinks this is profitable to fold and that + // we can fold it without inducing cycles in the graph. + if (!IsProfitableToFold(N, NodeStack[NodeStack.size()-2].getNode(), + NodeToMatch) || + !IsLegalToFold(N, NodeStack[NodeStack.size()-2].getNode(), + NodeToMatch, true/*We validate our own chains*/)) + break; + + continue; + } + case OPC_EmitInteger: { + MVT::SimpleValueType VT = + (MVT::SimpleValueType)MatcherTable[MatcherIndex++]; + int64_t Val = MatcherTable[MatcherIndex++]; + if (Val & 128) + Val = GetVBR(Val, MatcherTable, MatcherIndex); + RecordedNodes.push_back(CurDAG->getTargetConstant(Val, VT)); + continue; + } + case OPC_EmitRegister: { + MVT::SimpleValueType VT = + (MVT::SimpleValueType)MatcherTable[MatcherIndex++]; + unsigned RegNo = MatcherTable[MatcherIndex++]; + RecordedNodes.push_back(CurDAG->getRegister(RegNo, VT)); + 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]; + + if (Imm->getOpcode() == ISD::Constant) { + int64_t Val = cast<ConstantSDNode>(Imm)->getZExtValue(); + Imm = CurDAG->getTargetConstant(Val, Imm.getValueType()); + } else if (Imm->getOpcode() == ISD::ConstantFP) { + const ConstantFP *Val=cast<ConstantFPSDNode>(Imm)->getConstantFPValue(); + Imm = CurDAG->getTargetConstantFP(*Val, Imm.getValueType()); + } + + RecordedNodes.push_back(Imm); + continue; + } + + case OPC_EmitMergeInputChains: { + assert(InputChain.getNode() == 0 && + "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 + // together. However, if any of the input chains is actually one of the + // nodes matched in this pattern, then we have an intra-match reference. + // Ignore these because the newly token factored chain should not refer to + // the old nodes. + unsigned NumChains = MatcherTable[MatcherIndex++]; + assert(NumChains != 0 && "Can't TF zero chains"); + + assert(ChainNodesMatched.empty() && + "Should only have one EmitMergeInputChains per match"); + + // Read all of the chained nodes. + for (unsigned i = 0; i != NumChains; ++i) { + unsigned RecNo = MatcherTable[MatcherIndex++]; + assert(RecNo < RecordedNodes.size() && "Invalid CheckSame"); + ChainNodesMatched.push_back(RecordedNodes[RecNo].getNode()); + + // FIXME: What if other value results of the node have uses not matched + // by this pattern? + if (ChainNodesMatched.back() != NodeToMatch && + !RecordedNodes[RecNo].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); + 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)); + continue; + } + + case OPC_EmitNode: + case OPC_MorphNodeTo: { + uint16_t TargetOpc = MatcherTable[MatcherIndex++]; + TargetOpc |= (unsigned short)MatcherTable[MatcherIndex++] << 8; + unsigned EmitNodeInfo = MatcherTable[MatcherIndex++]; + // Get the result VT list. + unsigned NumVTs = MatcherTable[MatcherIndex++]; + SmallVector<EVT, 4> VTs; + for (unsigned i = 0; i != NumVTs; ++i) { + MVT::SimpleValueType VT = + (MVT::SimpleValueType)MatcherTable[MatcherIndex++]; + 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); + + // This is hot code, so optimize the two most common cases of 1 and 2 + // results. + SDVTList VTList; + if (VTs.size() == 1) + VTList = CurDAG->getVTList(VTs[0]); + else if (VTs.size() == 2) + VTList = CurDAG->getVTList(VTs[0], VTs[1]); + else + VTList = CurDAG->getVTList(VTs.data(), VTs.size()); + + // Get the operand list. + unsigned NumOps = MatcherTable[MatcherIndex++]; + SmallVector<SDValue, 8> Ops; + for (unsigned i = 0; i != NumOps; ++i) { + unsigned RecNo = MatcherTable[MatcherIndex++]; + if (RecNo & 128) + RecNo = GetVBR(RecNo, MatcherTable, MatcherIndex); + + assert(RecNo < RecordedNodes.size() && "Invalid EmitNode"); + Ops.push_back(RecordedNodes[RecNo]); + } + + // If there are variadic operands to add, handle them now. + if (EmitNodeInfo & OPFL_VariadicInfo) { + // Determine the start index to copy from. + unsigned FirstOpToCopy = getNumFixedFromVariadicInfo(EmitNodeInfo); + FirstOpToCopy += (EmitNodeInfo & OPFL_Chain) ? 1 : 0; + assert(NodeToMatch->getNumOperands() >= FirstOpToCopy && + "Invalid variadic node"); + // Copy all of the variadic operands, not including a potential flag + // input. + for (unsigned i = FirstOpToCopy, e = NodeToMatch->getNumOperands(); + i != e; ++i) { + SDValue V = NodeToMatch->getOperand(i); + if (V.getValueType() == MVT::Flag) break; + Ops.push_back(V); + } + } + + // If this has chain/flag inputs, add them. + if (EmitNodeInfo & OPFL_Chain) + Ops.push_back(InputChain); + if ((EmitNodeInfo & OPFL_FlagInput) && InputFlag.getNode() != 0) + Ops.push_back(InputFlag); + + // Create the node. + SDNode *Res = 0; + 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()); + + // Add all the non-flag/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)); + } + + } 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 (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 + // 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) { + MachineSDNode::mmo_iterator MemRefs = + MF->allocateMemRefsArray(MatchedMemRefs.size()); + std::copy(MatchedMemRefs.begin(), MatchedMemRefs.end(), MemRefs); + cast<MachineSDNode>(Res) + ->setMemRefs(MemRefs, MemRefs + MatchedMemRefs.size()); + } + + 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); + return Res; + } + + continue; + } + + case OPC_MarkFlagResults: { + unsigned NumNodes = MatcherTable[MatcherIndex++]; + + // Read and remember all the flag-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()); + } + 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 + // created nodes. + unsigned NumResults = MatcherTable[MatcherIndex++]; + + for (unsigned i = 0; i != NumResults; ++i) { + unsigned ResSlot = MatcherTable[MatcherIndex++]; + if (ResSlot & 128) + ResSlot = GetVBR(ResSlot, MatcherTable, MatcherIndex); + + assert(ResSlot < RecordedNodes.size() && "Invalid CheckSame"); + SDValue Res = RecordedNodes[ResSlot]; + + // FIXME2: Eliminate this horrible hack by fixing the 'Gen' program + // after (parallel) on input patterns are removed. This would also + // allow us to stop encoding #results in OPC_CompleteMatch's table + // entry. + if (NodeToMatch->getNumValues() <= i || + NodeToMatch->getValueType(i) == MVT::Other || + NodeToMatch->getValueType(i) == MVT::Flag) + break; + assert((NodeToMatch->getValueType(i) == Res.getValueType() || + NodeToMatch->getValueType(i) == MVT::iPTR || + Res.getValueType() == MVT::iPTR || + NodeToMatch->getValueType(i).getSizeInBits() == + Res.getValueType().getSizeInBits()) && + "invalid replacement"); + 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); + + 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. + while (1) { + if (MatchScopes.empty()) { + CannotYetSelect(NodeToMatch); + return 0; + } + + // Restore the interpreter state back to the point where the scope was + // formed. + MatchScope &LastScope = MatchScopes.back(); + RecordedNodes.resize(LastScope.NumRecordedNodes); + NodeStack.clear(); + NodeStack.append(LastScope.NodeStack.begin(), LastScope.NodeStack.end()); + N = NodeStack.back(); + + DEBUG(errs() << " Match failed at index " << MatcherIndex + << " continuing at " << LastScope.FailIndex << "\n"); + + if (LastScope.NumMatchedMemRefs != MatchedMemRefs.size()) + MatchedMemRefs.resize(LastScope.NumMatchedMemRefs); + MatcherIndex = LastScope.FailIndex; + + InputChain = LastScope.InputChain; + InputFlag = LastScope.InputFlag; + if (!LastScope.HasChainNodesMatched) + ChainNodesMatched.clear(); + if (!LastScope.HasFlagResultNodesMatched) + FlagResultNodesMatched.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 + // in the current scope to try. + unsigned NumToSkip = MatcherTable[MatcherIndex++]; + if (NumToSkip & 128) + NumToSkip = GetVBR(NumToSkip, MatcherTable, MatcherIndex); + + // If we have another child in this scope to match, update FailIndex and + // try it. + if (NumToSkip != 0) { + 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) { + if (N->getOpcode() == ISD::INTRINSIC_W_CHAIN || + N->getOpcode() == ISD::INTRINSIC_WO_CHAIN || + N->getOpcode() == ISD::INTRINSIC_VOID) + return CannotYetSelectIntrinsic(N); + std::string msg; raw_string_ostream Msg(msg); Msg << "Cannot yet select: "; diff --git a/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/lib/CodeGen/SelectionDAG/TargetLowering.cpp index e88af4f..0e54ca4 100644 --- a/lib/CodeGen/SelectionDAG/TargetLowering.cpp +++ b/lib/CodeGen/SelectionDAG/TargetLowering.cpp @@ -1441,8 +1441,10 @@ bool TargetLowering::SimplifyDemandedBits(SDValue Op, case ISD::TRUNCATE: { // Simplify the input, using demanded bit information, and compute the known // zero/one bits live out. + unsigned OperandBitWidth = + Op.getOperand(0).getValueType().getScalarType().getSizeInBits(); APInt TruncMask = NewMask; - TruncMask.zext(Op.getOperand(0).getValueSizeInBits()); + TruncMask.zext(OperandBitWidth); if (SimplifyDemandedBits(Op.getOperand(0), TruncMask, KnownZero, KnownOne, TLO, Depth+1)) return true; @@ -1453,15 +1455,14 @@ bool TargetLowering::SimplifyDemandedBits(SDValue Op, // on the known demanded bits. if (Op.getOperand(0).getNode()->hasOneUse()) { SDValue In = Op.getOperand(0); - unsigned InBitWidth = In.getValueSizeInBits(); switch (In.getOpcode()) { default: break; case ISD::SRL: // Shrink SRL by a constant if none of the high bits shifted in are // demanded. if (ConstantSDNode *ShAmt = dyn_cast<ConstantSDNode>(In.getOperand(1))){ - APInt HighBits = APInt::getHighBitsSet(InBitWidth, - InBitWidth - BitWidth); + APInt HighBits = APInt::getHighBitsSet(OperandBitWidth, + OperandBitWidth - BitWidth); HighBits = HighBits.lshr(ShAmt->getZExtValue()); HighBits.trunc(BitWidth); @@ -1607,7 +1608,7 @@ static bool ValueHasExactlyOneBitSet(SDValue Val, const SelectionDAG &DAG) { // Fall back to ComputeMaskedBits to catch other known cases. EVT OpVT = Val.getValueType(); - unsigned BitWidth = OpVT.getSizeInBits(); + unsigned BitWidth = OpVT.getScalarType().getSizeInBits(); APInt Mask = APInt::getAllOnesValue(BitWidth); APInt KnownZero, KnownOne; DAG.ComputeMaskedBits(Val, Mask, KnownZero, KnownOne); @@ -1775,7 +1776,7 @@ TargetLowering::SimplifySetCC(EVT VT, SDValue N0, SDValue N1, break; // todo, be more careful with signed comparisons } } else if (N0.getOpcode() == ISD::SIGN_EXTEND_INREG && - (Cond == ISD::SETEQ || Cond == ISD::SETNE)) { + (Cond == ISD::SETEQ || Cond == ISD::SETNE)) { EVT ExtSrcTy = cast<VTSDNode>(N0.getOperand(1))->getVT(); unsigned ExtSrcTyBits = ExtSrcTy.getSizeInBits(); EVT ExtDstTy = N0.getValueType(); @@ -1809,22 +1810,21 @@ TargetLowering::SimplifySetCC(EVT VT, SDValue N0, SDValue N1, Cond); } else if ((N1C->isNullValue() || N1C->getAPIntValue() == 1) && (Cond == ISD::SETEQ || Cond == ISD::SETNE)) { - // SETCC (SETCC), [0|1], [EQ|NE] -> SETCC - if (N0.getOpcode() == ISD::SETCC) { + if (N0.getOpcode() == ISD::SETCC && + isTypeLegal(VT) && VT.bitsLE(N0.getValueType())) { bool TrueWhenTrue = (Cond == ISD::SETEQ) ^ (N1C->getAPIntValue() != 1); if (TrueWhenTrue) - return N0; - + return DAG.getNode(ISD::TRUNCATE, dl, VT, N0); // Invert the condition. ISD::CondCode CC = cast<CondCodeSDNode>(N0.getOperand(2))->get(); CC = ISD::getSetCCInverse(CC, N0.getOperand(0).getValueType().isInteger()); return DAG.getSetCC(dl, VT, N0.getOperand(0), N0.getOperand(1), CC); } - + if ((N0.getOpcode() == ISD::XOR || - (N0.getOpcode() == ISD::AND && + (N0.getOpcode() == ISD::AND && N0.getOperand(0).getOpcode() == ISD::XOR && N0.getOperand(1) == N0.getOperand(0).getOperand(1))) && isa<ConstantSDNode>(N0.getOperand(1)) && @@ -1847,9 +1847,36 @@ TargetLowering::SimplifySetCC(EVT VT, SDValue N0, SDValue N1, N0.getOperand(0).getOperand(0), N0.getOperand(1)); } + return DAG.getSetCC(dl, VT, Val, N1, Cond == ISD::SETEQ ? ISD::SETNE : ISD::SETEQ); } + } else if (N1C->getAPIntValue() == 1 && + (VT == MVT::i1 || + getBooleanContents() == ZeroOrOneBooleanContent)) { + SDValue Op0 = N0; + if (Op0.getOpcode() == ISD::TRUNCATE) + Op0 = Op0.getOperand(0); + + if ((Op0.getOpcode() == ISD::XOR) && + Op0.getOperand(0).getOpcode() == ISD::SETCC && + Op0.getOperand(1).getOpcode() == ISD::SETCC) { + // (xor (setcc), (setcc)) == / != 1 -> (setcc) != / == (setcc) + Cond = (Cond == ISD::SETEQ) ? ISD::SETNE : ISD::SETEQ; + return DAG.getSetCC(dl, VT, Op0.getOperand(0), Op0.getOperand(1), + Cond); + } else if (Op0.getOpcode() == ISD::AND && + isa<ConstantSDNode>(Op0.getOperand(1)) && + cast<ConstantSDNode>(Op0.getOperand(1))->getAPIntValue() == 1) { + // If this is (X&1) == / != 1, normalize it to (X&1) != / == 0. + if (Op0.getValueType() != VT) + Op0 = DAG.getNode(ISD::AND, dl, VT, + DAG.getNode(ISD::TRUNCATE, dl, VT, Op0.getOperand(0)), + DAG.getConstant(1, VT)); + return DAG.getSetCC(dl, VT, Op0, + DAG.getConstant(0, Op0.getValueType()), + Cond == ISD::SETEQ ? ISD::SETNE : ISD::SETEQ); + } } } diff --git a/lib/CodeGen/SimpleRegisterCoalescing.cpp b/lib/CodeGen/SimpleRegisterCoalescing.cpp index e7b0cff..ce72b2f 100644 --- a/lib/CodeGen/SimpleRegisterCoalescing.cpp +++ b/lib/CodeGen/SimpleRegisterCoalescing.cpp @@ -662,7 +662,7 @@ bool SimpleRegisterCoalescing::ReMaterializeTrivialDef(LiveInterval &SrcInt, if (!tii_->isTriviallyReMaterializable(DefMI, AA)) return false; bool SawStore = false; - if (!DefMI->isSafeToMove(tii_, SawStore, AA)) + if (!DefMI->isSafeToMove(tii_, AA, SawStore)) return false; if (TID.getNumDefs() != 1) return false; @@ -702,7 +702,8 @@ bool SimpleRegisterCoalescing::ReMaterializeTrivialDef(LiveInterval &SrcInt, for (const unsigned* SR = tri_->getSubRegisters(DstReg); *SR; ++SR) { if (!li_->hasInterval(*SR)) continue; - DLR = li_->getInterval(*SR).getLiveRangeContaining(DefIdx); + const LiveRange *DLR = + li_->getInterval(*SR).getLiveRangeContaining(DefIdx); if (DLR && DLR->valno->getCopy() == CopyMI) DLR->valno->setCopy(0); } @@ -741,9 +742,21 @@ bool SimpleRegisterCoalescing::ReMaterializeTrivialDef(LiveInterval &SrcInt, NewMI->addOperand(MO); if (MO.isDef() && li_->hasInterval(MO.getReg())) { unsigned Reg = MO.getReg(); - DLR = li_->getInterval(Reg).getLiveRangeContaining(DefIdx); + const LiveRange *DLR = + li_->getInterval(Reg).getLiveRangeContaining(DefIdx); if (DLR && DLR->valno->getCopy() == CopyMI) DLR->valno->setCopy(0); + // Handle subregs as well + if (TargetRegisterInfo::isPhysicalRegister(Reg)) { + for (const unsigned* SR = tri_->getSubRegisters(Reg); *SR; ++SR) { + if (!li_->hasInterval(*SR)) + continue; + const LiveRange *DLR = + li_->getInterval(*SR).getLiveRangeContaining(DefIdx); + if (DLR && DLR->valno->getCopy() == CopyMI) + DLR->valno->setCopy(0); + } + } } } @@ -752,6 +765,7 @@ bool SimpleRegisterCoalescing::ReMaterializeTrivialDef(LiveInterval &SrcInt, CopyMI->eraseFromParent(); ReMatCopies.insert(CopyMI); ReMatDefs.insert(DefMI); + DEBUG(dbgs() << "Remat: " << *NewMI); ++NumReMats; return true; } @@ -1705,6 +1719,7 @@ bool SimpleRegisterCoalescing::JoinCopy(CopyRec &TheCopy, bool &Again) { (AdjustCopiesBackFrom(SrcInt, DstInt, CopyMI) || RemoveCopyByCommutingDef(SrcInt, DstInt, CopyMI))) { JoinedCopies.insert(CopyMI); + DEBUG(dbgs() << "Trivial!\n"); return true; } @@ -1864,7 +1879,7 @@ static unsigned ComputeUltimateVN(VNInfo *VNI, // If the VN has already been computed, just return it. if (ThisValNoAssignments[VN] >= 0) return ThisValNoAssignments[VN]; -// assert(ThisValNoAssignments[VN] != -2 && "Cyclic case?"); + assert(ThisValNoAssignments[VN] != -2 && "Cyclic value numbers"); // If this val is not a copy from the other val, then it must be a new value // number in the destination. diff --git a/lib/CodeGen/TargetInstrInfoImpl.cpp b/lib/CodeGen/TargetInstrInfoImpl.cpp index a0fccab..e9e998f 100644 --- a/lib/CodeGen/TargetInstrInfoImpl.cpp +++ b/lib/CodeGen/TargetInstrInfoImpl.cpp @@ -150,6 +150,11 @@ void TargetInstrInfoImpl::reMaterialize(MachineBasicBlock &MBB, MBB.insert(I, MI); } +bool TargetInstrInfoImpl::produceSameValue(const MachineInstr *MI0, + const MachineInstr *MI1) const { + return MI0->isIdenticalTo(MI1, MachineInstr::IgnoreVRegDefs); +} + MachineInstr *TargetInstrInfoImpl::duplicate(MachineInstr *Orig, MachineFunction &MF) const { assert(!Orig->getDesc().isNotDuplicable() && @@ -157,37 +162,6 @@ MachineInstr *TargetInstrInfoImpl::duplicate(MachineInstr *Orig, return MF.CloneMachineInstr(Orig); } -bool -TargetInstrInfoImpl::isIdentical(const MachineInstr *MI, - const MachineInstr *Other, - const MachineRegisterInfo *MRI) const { - if (MI->getOpcode() != Other->getOpcode() || - MI->getNumOperands() != Other->getNumOperands()) - return false; - - for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) { - const MachineOperand &MO = MI->getOperand(i); - const MachineOperand &OMO = Other->getOperand(i); - if (MO.isReg() && MO.isDef()) { - assert(OMO.isReg() && OMO.isDef()); - unsigned Reg = MO.getReg(); - if (TargetRegisterInfo::isPhysicalRegister(Reg)) { - if (Reg != OMO.getReg()) - return false; - } else if (MRI->getRegClass(MO.getReg()) != - MRI->getRegClass(OMO.getReg())) - return false; - - continue; - } - - if (!MO.isIdenticalTo(OMO)) - return false; - } - - return true; -} - unsigned TargetInstrInfoImpl::GetFunctionSizeInBytes(const MachineFunction &MF) const { unsigned FnSize = 0; diff --git a/lib/CodeGen/TargetLoweringObjectFileImpl.cpp b/lib/CodeGen/TargetLoweringObjectFileImpl.cpp index 190b533..ef6e129 100644 --- a/lib/CodeGen/TargetLoweringObjectFileImpl.cpp +++ b/lib/CodeGen/TargetLoweringObjectFileImpl.cpp @@ -33,6 +33,7 @@ #include "llvm/ADT/SmallString.h" #include "llvm/ADT/StringExtras.h" using namespace llvm; +using namespace dwarf; //===----------------------------------------------------------------------===// // ELF @@ -549,8 +550,8 @@ void TargetLoweringObjectFileMachO::Initialize(MCContext &Ctx, } // Exception Handling. - LSDASection = getMachOSection("__DATA", "__gcc_except_tab", 0, - SectionKind::getDataRel()); + LSDASection = getMachOSection("__TEXT", "__gcc_except_tab", 0, + SectionKind::getReadOnlyWithRel()); EHFrameSection = getMachOSection("__TEXT", "__eh_frame", MCSectionMachO::S_COALESCED | @@ -738,11 +739,23 @@ getSymbolForDwarfGlobalReference(const GlobalValue *GV, Mangler *Mang, MachineModuleInfo *MMI, unsigned Encoding) const { // The mach-o version of this method defaults to returning a stub reference. - if (Encoding & dwarf::DW_EH_PE_indirect) { + if (Encoding & DW_EH_PE_indirect) { + MachineModuleInfoMachO &MachOMMI = + MMI->getObjFileInfo<MachineModuleInfoMachO>(); + SmallString<128> Name; Mang->getNameWithPrefix(Name, GV, true); Name += "$non_lazy_ptr"; + + // Add information about the stub reference to MachOMMI so that the stub + // gets emitted by the asmprinter. MCSymbol *Sym = getContext().GetOrCreateSymbol(Name.str()); + MCSymbol *&StubSym = MachOMMI.getGVStubEntry(Sym); + if (StubSym == 0) { + Name.clear(); + Mang->getNameWithPrefix(Name, GV, false); + StubSym = getContext().GetOrCreateSymbol(Name.str()); + } return TargetLoweringObjectFile:: getSymbolForDwarfReference(Sym, MMI, @@ -753,6 +766,21 @@ getSymbolForDwarfGlobalReference(const GlobalValue *GV, Mangler *Mang, getSymbolForDwarfGlobalReference(GV, Mang, MMI, Encoding); } +unsigned TargetLoweringObjectFileMachO::getPersonalityEncoding() const { + return DW_EH_PE_indirect | DW_EH_PE_pcrel | DW_EH_PE_sdata4; +} + +unsigned TargetLoweringObjectFileMachO::getLSDAEncoding() const { + return DW_EH_PE_pcrel; +} + +unsigned TargetLoweringObjectFileMachO::getFDEEncoding() const { + return DW_EH_PE_pcrel; +} + +unsigned TargetLoweringObjectFileMachO::getTTypeEncoding() const { + return DW_EH_PE_indirect | DW_EH_PE_pcrel | DW_EH_PE_sdata4; +} //===----------------------------------------------------------------------===// // COFF diff --git a/lib/CodeGen/TwoAddressInstructionPass.cpp b/lib/CodeGen/TwoAddressInstructionPass.cpp index 6f4ca82..0ba3843 100644 --- a/lib/CodeGen/TwoAddressInstructionPass.cpp +++ b/lib/CodeGen/TwoAddressInstructionPass.cpp @@ -160,7 +160,7 @@ bool TwoAddressInstructionPass::Sink3AddrInstruction(MachineBasicBlock *MBB, MachineBasicBlock::iterator OldPos) { // Check if it's safe to move this instruction. bool SeenStore = true; // Be conservative. - if (!MI->isSafeToMove(TII, SeenStore, AA)) + if (!MI->isSafeToMove(TII, AA, SeenStore)) return false; unsigned DefReg = 0; @@ -1028,7 +1028,7 @@ bool TwoAddressInstructionPass::runOnMachineFunction(MachineFunction &MF) { // copying it. if (DefMI && DefMI->getDesc().isAsCheapAsAMove() && - DefMI->isSafeToReMat(TII, regB, AA) && + DefMI->isSafeToReMat(TII, AA, regB) && isProfitableToReMat(regB, rc, mi, DefMI, mbbi, Dist)){ DEBUG(dbgs() << "2addr: REMATTING : " << *DefMI << "\n"); unsigned regASubIdx = mi->getOperand(DstIdx).getSubReg(); diff --git a/lib/CodeGen/VirtRegMap.cpp b/lib/CodeGen/VirtRegMap.cpp index 5956b61..ed02696 100644 --- a/lib/CodeGen/VirtRegMap.cpp +++ b/lib/CodeGen/VirtRegMap.cpp @@ -261,19 +261,21 @@ bool VirtRegMap::FindUnusedRegisters(LiveIntervals* LIs) { void VirtRegMap::print(raw_ostream &OS, const Module* M) const { const TargetRegisterInfo* TRI = MF->getTarget().getRegisterInfo(); + const MachineRegisterInfo &MRI = MF->getRegInfo(); OS << "********** REGISTER MAP **********\n"; for (unsigned i = TargetRegisterInfo::FirstVirtualRegister, e = MF->getRegInfo().getLastVirtReg(); i <= e; ++i) { if (Virt2PhysMap[i] != (unsigned)VirtRegMap::NO_PHYS_REG) OS << "[reg" << i << " -> " << TRI->getName(Virt2PhysMap[i]) - << "]\n"; + << "] " << MRI.getRegClass(i)->getName() << "\n"; } for (unsigned i = TargetRegisterInfo::FirstVirtualRegister, e = MF->getRegInfo().getLastVirtReg(); i <= e; ++i) if (Virt2StackSlotMap[i] != VirtRegMap::NO_STACK_SLOT) - OS << "[reg" << i << " -> fi#" << Virt2StackSlotMap[i] << "]\n"; + OS << "[reg" << i << " -> fi#" << Virt2StackSlotMap[i] + << "] " << MRI.getRegClass(i)->getName() << "\n"; OS << '\n'; } diff --git a/lib/CodeGen/VirtRegRewriter.cpp b/lib/CodeGen/VirtRegRewriter.cpp index 84e0398..7aa0a91 100644 --- a/lib/CodeGen/VirtRegRewriter.cpp +++ b/lib/CodeGen/VirtRegRewriter.cpp @@ -46,7 +46,7 @@ namespace { static cl::opt<RewriterName> RewriterOpt("rewriter", - cl::desc("Rewriter to use: (default: local)"), + cl::desc("Rewriter to use (default=local)"), cl::Prefix, cl::values(clEnumVal(local, "local rewriter"), clEnumVal(trivial, "trivial rewriter"), |
