| //===-- PhyRegAlloc.cpp ---------------------------------------------------===// |
| // |
| // The LLVM Compiler Infrastructure |
| // |
| // This file was developed by the LLVM research group and is distributed under |
| // the University of Illinois Open Source License. See LICENSE.TXT for details. |
| // |
| //===----------------------------------------------------------------------===// |
| // |
| // Traditional graph-coloring global register allocator currently used |
| // by the SPARC back-end. |
| // |
| // NOTE: This register allocator has some special support |
| // for the Reoptimizer, such as not saving some registers on calls to |
| // the first-level instrumentation function. |
| // |
| // NOTE 2: This register allocator can save its state in a global |
| // variable in the module it's working on. This feature is not |
| // thread-safe; if you have doubts, leave it turned off. |
| // |
| //===----------------------------------------------------------------------===// |
| |
| #include "AllocInfo.h" |
| #include "IGNode.h" |
| #include "PhyRegAlloc.h" |
| #include "RegAllocCommon.h" |
| #include "RegClass.h" |
| #include "llvm/Constants.h" |
| #include "llvm/DerivedTypes.h" |
| #include "llvm/iOther.h" |
| #include "llvm/Module.h" |
| #include "llvm/Type.h" |
| #include "llvm/Analysis/LoopInfo.h" |
| #include "llvm/CodeGen/FunctionLiveVarInfo.h" |
| #include "llvm/CodeGen/InstrSelection.h" |
| #include "llvm/CodeGen/MachineCodeForInstruction.h" |
| #include "llvm/CodeGen/MachineFunction.h" |
| #include "llvm/CodeGen/MachineFunctionInfo.h" |
| #include "llvm/CodeGen/MachineInstr.h" |
| #include "llvm/CodeGen/MachineInstrBuilder.h" |
| #include "llvm/CodeGen/MachineInstrAnnot.h" |
| #include "llvm/CodeGen/Passes.h" |
| #include "llvm/Support/InstIterator.h" |
| #include "llvm/Target/TargetInstrInfo.h" |
| #include "Support/CommandLine.h" |
| #include "Support/SetOperations.h" |
| #include "Support/STLExtras.h" |
| #include <cmath> |
| |
| namespace llvm { |
| |
| RegAllocDebugLevel_t DEBUG_RA; |
| |
| /// The reoptimizer wants to be able to grovel through the register |
| /// allocator's state after it has done its job. This is a hack. |
| /// |
| PhyRegAlloc::SavedStateMapTy ExportedFnAllocState; |
| const bool SaveStateToModule = true; |
| |
| static cl::opt<RegAllocDebugLevel_t, true> |
| DRA_opt("dregalloc", cl::Hidden, cl::location(DEBUG_RA), |
| cl::desc("enable register allocation debugging information"), |
| cl::values( |
| clEnumValN(RA_DEBUG_None , "n", "disable debug output"), |
| clEnumValN(RA_DEBUG_Results, "y", "debug output for allocation results"), |
| clEnumValN(RA_DEBUG_Coloring, "c", "debug output for graph coloring step"), |
| clEnumValN(RA_DEBUG_Interference,"ig","debug output for interference graphs"), |
| clEnumValN(RA_DEBUG_LiveRanges , "lr","debug output for live ranges"), |
| clEnumValN(RA_DEBUG_Verbose, "v", "extra debug output"), |
| 0)); |
| |
| static cl::opt<bool> |
| SaveRegAllocState("save-ra-state", cl::Hidden, |
| cl::desc("write reg. allocator state into module")); |
| |
| FunctionPass *getRegisterAllocator(TargetMachine &T) { |
| return new PhyRegAlloc (T); |
| } |
| |
| void PhyRegAlloc::getAnalysisUsage(AnalysisUsage &AU) const { |
| AU.addRequired<LoopInfo> (); |
| AU.addRequired<FunctionLiveVarInfo> (); |
| } |
| |
| |
| /// Initialize interference graphs (one in each reg class) and IGNodeLists |
| /// (one in each IG). The actual nodes will be pushed later. |
| /// |
| void PhyRegAlloc::createIGNodeListsAndIGs() { |
| if (DEBUG_RA >= RA_DEBUG_LiveRanges) std::cerr << "Creating LR lists ...\n"; |
| |
| LiveRangeMapType::const_iterator HMI = LRI->getLiveRangeMap()->begin(); |
| LiveRangeMapType::const_iterator HMIEnd = LRI->getLiveRangeMap()->end(); |
| |
| for (; HMI != HMIEnd ; ++HMI ) { |
| if (HMI->first) { |
| LiveRange *L = HMI->second; // get the LiveRange |
| if (!L) { |
| if (DEBUG_RA) |
| std::cerr << "\n**** ?!?WARNING: NULL LIVE RANGE FOUND FOR: " |
| << RAV(HMI->first) << "****\n"; |
| continue; |
| } |
| |
| // if the Value * is not null, and LR is not yet written to the IGNodeList |
| if (!(L->getUserIGNode()) ) { |
| RegClass *const RC = // RegClass of first value in the LR |
| RegClassList[ L->getRegClassID() ]; |
| RC->addLRToIG(L); // add this LR to an IG |
| } |
| } |
| } |
| |
| // init RegClassList |
| for ( unsigned rc=0; rc < NumOfRegClasses ; rc++) |
| RegClassList[rc]->createInterferenceGraph(); |
| |
| if (DEBUG_RA >= RA_DEBUG_LiveRanges) std::cerr << "LRLists Created!\n"; |
| } |
| |
| |
| /// Add all interferences for a given instruction. Interference occurs only |
| /// if the LR of Def (Inst or Arg) is of the same reg class as that of live |
| /// var. The live var passed to this function is the LVset AFTER the |
| /// instruction. |
| /// |
| void PhyRegAlloc::addInterference(const Value *Def, const ValueSet *LVSet, |
| bool isCallInst) { |
| ValueSet::const_iterator LIt = LVSet->begin(); |
| |
| // get the live range of instruction |
| const LiveRange *const LROfDef = LRI->getLiveRangeForValue( Def ); |
| |
| IGNode *const IGNodeOfDef = LROfDef->getUserIGNode(); |
| assert( IGNodeOfDef ); |
| |
| RegClass *const RCOfDef = LROfDef->getRegClass(); |
| |
| // for each live var in live variable set |
| for ( ; LIt != LVSet->end(); ++LIt) { |
| |
| if (DEBUG_RA >= RA_DEBUG_Verbose) |
| std::cerr << "< Def=" << RAV(Def) << ", Lvar=" << RAV(*LIt) << "> "; |
| |
| // get the live range corresponding to live var |
| LiveRange *LROfVar = LRI->getLiveRangeForValue(*LIt); |
| |
| // LROfVar can be null if it is a const since a const |
| // doesn't have a dominating def - see Assumptions above |
| if (LROfVar) |
| if (LROfDef != LROfVar) // do not set interf for same LR |
| if (RCOfDef == LROfVar->getRegClass()) // 2 reg classes are the same |
| RCOfDef->setInterference( LROfDef, LROfVar); |
| } |
| } |
| |
| |
| /// For a call instruction, this method sets the CallInterference flag in |
| /// the LR of each variable live in the Live Variable Set live after the |
| /// call instruction (except the return value of the call instruction - since |
| /// the return value does not interfere with that call itself). |
| /// |
| void PhyRegAlloc::setCallInterferences(const MachineInstr *MInst, |
| const ValueSet *LVSetAft) { |
| if (DEBUG_RA >= RA_DEBUG_Interference) |
| std::cerr << "\n For call inst: " << *MInst; |
| |
| // for each live var in live variable set after machine inst |
| for (ValueSet::const_iterator LIt = LVSetAft->begin(), LEnd = LVSetAft->end(); |
| LIt != LEnd; ++LIt) { |
| |
| // get the live range corresponding to live var |
| LiveRange *const LR = LRI->getLiveRangeForValue(*LIt ); |
| |
| // LR can be null if it is a const since a const |
| // doesn't have a dominating def - see Assumptions above |
| if (LR ) { |
| if (DEBUG_RA >= RA_DEBUG_Interference) { |
| std::cerr << "\n\tLR after Call: "; |
| printSet(*LR); |
| } |
| LR->setCallInterference(); |
| if (DEBUG_RA >= RA_DEBUG_Interference) { |
| std::cerr << "\n ++After adding call interference for LR: " ; |
| printSet(*LR); |
| } |
| } |
| |
| } |
| |
| // Now find the LR of the return value of the call |
| // We do this because, we look at the LV set *after* the instruction |
| // to determine, which LRs must be saved across calls. The return value |
| // of the call is live in this set - but it does not interfere with call |
| // (i.e., we can allocate a volatile register to the return value) |
| CallArgsDescriptor* argDesc = CallArgsDescriptor::get(MInst); |
| |
| if (const Value *RetVal = argDesc->getReturnValue()) { |
| LiveRange *RetValLR = LRI->getLiveRangeForValue( RetVal ); |
| assert( RetValLR && "No LR for RetValue of call"); |
| RetValLR->clearCallInterference(); |
| } |
| |
| // If the CALL is an indirect call, find the LR of the function pointer. |
| // That has a call interference because it conflicts with outgoing args. |
| if (const Value *AddrVal = argDesc->getIndirectFuncPtr()) { |
| LiveRange *AddrValLR = LRI->getLiveRangeForValue( AddrVal ); |
| assert( AddrValLR && "No LR for indirect addr val of call"); |
| AddrValLR->setCallInterference(); |
| } |
| } |
| |
| |
| /// Create interferences in the IG of each RegClass, and calculate the spill |
| /// cost of each Live Range (it is done in this method to save another pass |
| /// over the code). |
| /// |
| void PhyRegAlloc::buildInterferenceGraphs() { |
| if (DEBUG_RA >= RA_DEBUG_Interference) |
| std::cerr << "Creating interference graphs ...\n"; |
| |
| unsigned BBLoopDepthCost; |
| for (MachineFunction::iterator BBI = MF->begin(), BBE = MF->end(); |
| BBI != BBE; ++BBI) { |
| const MachineBasicBlock &MBB = *BBI; |
| const BasicBlock *BB = MBB.getBasicBlock(); |
| |
| // find the 10^(loop_depth) of this BB |
| BBLoopDepthCost = (unsigned)pow(10.0, LoopDepthCalc->getLoopDepth(BB)); |
| |
| // get the iterator for machine instructions |
| MachineBasicBlock::const_iterator MII = MBB.begin(); |
| |
| // iterate over all the machine instructions in BB |
| for ( ; MII != MBB.end(); ++MII) { |
| const MachineInstr *MInst = *MII; |
| |
| // get the LV set after the instruction |
| const ValueSet &LVSetAI = LVI->getLiveVarSetAfterMInst(MInst, BB); |
| bool isCallInst = TM.getInstrInfo().isCall(MInst->getOpCode()); |
| |
| if (isCallInst) { |
| // set the isCallInterference flag of each live range which extends |
| // across this call instruction. This information is used by graph |
| // coloring algorithm to avoid allocating volatile colors to live ranges |
| // that span across calls (since they have to be saved/restored) |
| setCallInterferences(MInst, &LVSetAI); |
| } |
| |
| // iterate over all MI operands to find defs |
| for (MachineInstr::const_val_op_iterator OpI = MInst->begin(), |
| OpE = MInst->end(); OpI != OpE; ++OpI) { |
| if (OpI.isDefOnly() || OpI.isDefAndUse()) // create a new LR since def |
| addInterference(*OpI, &LVSetAI, isCallInst); |
| |
| // Calculate the spill cost of each live range |
| LiveRange *LR = LRI->getLiveRangeForValue(*OpI); |
| if (LR) LR->addSpillCost(BBLoopDepthCost); |
| } |
| |
| // Mark all operands of pseudo-instructions as interfering with one |
| // another. This must be done because pseudo-instructions may be |
| // expanded to multiple instructions by the assembler, so all the |
| // operands must get distinct registers. |
| if (TM.getInstrInfo().isPseudoInstr(MInst->getOpCode())) |
| addInterf4PseudoInstr(MInst); |
| |
| // Also add interference for any implicit definitions in a machine |
| // instr (currently, only calls have this). |
| unsigned NumOfImpRefs = MInst->getNumImplicitRefs(); |
| for (unsigned z=0; z < NumOfImpRefs; z++) |
| if (MInst->getImplicitOp(z).opIsDefOnly() || |
| MInst->getImplicitOp(z).opIsDefAndUse()) |
| addInterference( MInst->getImplicitRef(z), &LVSetAI, isCallInst ); |
| |
| } // for all machine instructions in BB |
| } // for all BBs in function |
| |
| // add interferences for function arguments. Since there are no explicit |
| // defs in the function for args, we have to add them manually |
| addInterferencesForArgs(); |
| |
| if (DEBUG_RA >= RA_DEBUG_Interference) |
| std::cerr << "Interference graphs calculated!\n"; |
| } |
| |
| |
| /// Mark all operands of the given MachineInstr as interfering with one |
| /// another. |
| /// |
| void PhyRegAlloc::addInterf4PseudoInstr(const MachineInstr *MInst) { |
| bool setInterf = false; |
| |
| // iterate over MI operands to find defs |
| for (MachineInstr::const_val_op_iterator It1 = MInst->begin(), |
| ItE = MInst->end(); It1 != ItE; ++It1) { |
| const LiveRange *LROfOp1 = LRI->getLiveRangeForValue(*It1); |
| assert((LROfOp1 || !It1.isUseOnly())&&"No LR for Def in PSEUDO insruction"); |
| |
| MachineInstr::const_val_op_iterator It2 = It1; |
| for (++It2; It2 != ItE; ++It2) { |
| const LiveRange *LROfOp2 = LRI->getLiveRangeForValue(*It2); |
| |
| if (LROfOp2) { |
| RegClass *RCOfOp1 = LROfOp1->getRegClass(); |
| RegClass *RCOfOp2 = LROfOp2->getRegClass(); |
| |
| if (RCOfOp1 == RCOfOp2 ){ |
| RCOfOp1->setInterference( LROfOp1, LROfOp2 ); |
| setInterf = true; |
| } |
| } // if Op2 has a LR |
| } // for all other defs in machine instr |
| } // for all operands in an instruction |
| |
| if (!setInterf && MInst->getNumOperands() > 2) { |
| std::cerr << "\nInterf not set for any operand in pseudo instr:\n"; |
| std::cerr << *MInst; |
| assert(0 && "Interf not set for pseudo instr with > 2 operands" ); |
| } |
| } |
| |
| |
| /// Add interferences for incoming arguments to a function. |
| /// |
| void PhyRegAlloc::addInterferencesForArgs() { |
| // get the InSet of root BB |
| const ValueSet &InSet = LVI->getInSetOfBB(&Fn->front()); |
| |
| for (Function::const_aiterator AI = Fn->abegin(); AI != Fn->aend(); ++AI) { |
| // add interferences between args and LVars at start |
| addInterference(AI, &InSet, false); |
| |
| if (DEBUG_RA >= RA_DEBUG_Interference) |
| std::cerr << " - %% adding interference for argument " << RAV(AI) << "\n"; |
| } |
| } |
| |
| |
| /// The following are utility functions used solely by updateMachineCode and |
| /// the functions that it calls. They should probably be folded back into |
| /// updateMachineCode at some point. |
| /// |
| |
| // used by: updateMachineCode (1 time), PrependInstructions (1 time) |
| inline void InsertBefore(MachineInstr* newMI, MachineBasicBlock& MBB, |
| MachineBasicBlock::iterator& MII) { |
| MII = MBB.insert(MII, newMI); |
| ++MII; |
| } |
| |
| // used by: AppendInstructions (1 time) |
| inline void InsertAfter(MachineInstr* newMI, MachineBasicBlock& MBB, |
| MachineBasicBlock::iterator& MII) { |
| ++MII; // insert before the next instruction |
| MII = MBB.insert(MII, newMI); |
| } |
| |
| // used by: updateMachineCode (1 time) |
| inline void DeleteInstruction(MachineBasicBlock& MBB, |
| MachineBasicBlock::iterator& MII) { |
| MII = MBB.erase(MII); |
| } |
| |
| // used by: updateMachineCode (1 time) |
| inline void SubstituteInPlace(MachineInstr* newMI, MachineBasicBlock& MBB, |
| MachineBasicBlock::iterator MII) { |
| *MII = newMI; |
| } |
| |
| // used by: updateMachineCode (2 times) |
| inline void PrependInstructions(std::vector<MachineInstr *> &IBef, |
| MachineBasicBlock& MBB, |
| MachineBasicBlock::iterator& MII, |
| const std::string& msg) { |
| if (!IBef.empty()) { |
| MachineInstr* OrigMI = *MII; |
| std::vector<MachineInstr *>::iterator AdIt; |
| for (AdIt = IBef.begin(); AdIt != IBef.end() ; ++AdIt) { |
| if (DEBUG_RA) { |
| if (OrigMI) std::cerr << "For MInst:\n " << *OrigMI; |
| std::cerr << msg << "PREPENDed instr:\n " << **AdIt << "\n"; |
| } |
| InsertBefore(*AdIt, MBB, MII); |
| } |
| } |
| } |
| |
| // used by: updateMachineCode (1 time) |
| inline void AppendInstructions(std::vector<MachineInstr *> &IAft, |
| MachineBasicBlock& MBB, |
| MachineBasicBlock::iterator& MII, |
| const std::string& msg) { |
| if (!IAft.empty()) { |
| MachineInstr* OrigMI = *MII; |
| std::vector<MachineInstr *>::iterator AdIt; |
| for ( AdIt = IAft.begin(); AdIt != IAft.end() ; ++AdIt ) { |
| if (DEBUG_RA) { |
| if (OrigMI) std::cerr << "For MInst:\n " << *OrigMI; |
| std::cerr << msg << "APPENDed instr:\n " << **AdIt << "\n"; |
| } |
| InsertAfter(*AdIt, MBB, MII); |
| } |
| } |
| } |
| |
| /// Set the registers for operands in the given MachineInstr, if a register was |
| /// successfully allocated. Return true if any of its operands has been marked |
| /// for spill. |
| /// |
| bool PhyRegAlloc::markAllocatedRegs(MachineInstr* MInst) |
| { |
| bool instrNeedsSpills = false; |
| |
| // First, set the registers for operands in the machine instruction |
| // if a register was successfully allocated. Do this first because we |
| // will need to know which registers are already used by this instr'n. |
| for (unsigned OpNum=0; OpNum < MInst->getNumOperands(); ++OpNum) { |
| MachineOperand& Op = MInst->getOperand(OpNum); |
| if (Op.getType() == MachineOperand::MO_VirtualRegister || |
| Op.getType() == MachineOperand::MO_CCRegister) { |
| const Value *const Val = Op.getVRegValue(); |
| if (const LiveRange* LR = LRI->getLiveRangeForValue(Val)) { |
| // Remember if any operand needs spilling |
| instrNeedsSpills |= LR->isMarkedForSpill(); |
| |
| // An operand may have a color whether or not it needs spilling |
| if (LR->hasColor()) |
| MInst->SetRegForOperand(OpNum, |
| MRI.getUnifiedRegNum(LR->getRegClassID(), |
| LR->getColor())); |
| } |
| } |
| } // for each operand |
| |
| return instrNeedsSpills; |
| } |
| |
| /// Mark allocated registers (using markAllocatedRegs()) on the instruction |
| /// that MII points to. Then, if it's a call instruction, insert caller-saving |
| /// code before and after it. Finally, insert spill code before and after it, |
| /// using insertCode4SpilledLR(). |
| /// |
| void PhyRegAlloc::updateInstruction(MachineBasicBlock::iterator& MII, |
| MachineBasicBlock &MBB) { |
| MachineInstr* MInst = *MII; |
| unsigned Opcode = MInst->getOpCode(); |
| |
| // Reset tmp stack positions so they can be reused for each machine instr. |
| MF->getInfo()->popAllTempValues(); |
| |
| // Mark the operands for which regs have been allocated. |
| bool instrNeedsSpills = markAllocatedRegs(*MII); |
| |
| #ifndef NDEBUG |
| // Mark that the operands have been updated. Later, |
| // setRelRegsUsedByThisInst() is called to find registers used by each |
| // MachineInst, and it should not be used for an instruction until |
| // this is done. This flag just serves as a sanity check. |
| OperandsColoredMap[MInst] = true; |
| #endif |
| |
| // Now insert caller-saving code before/after the call. |
| // Do this before inserting spill code since some registers must be |
| // used by save/restore and spill code should not use those registers. |
| if (TM.getInstrInfo().isCall(Opcode)) { |
| AddedInstrns &AI = AddedInstrMap[MInst]; |
| insertCallerSavingCode(AI.InstrnsBefore, AI.InstrnsAfter, MInst, |
| MBB.getBasicBlock()); |
| } |
| |
| // Now insert spill code for remaining operands not allocated to |
| // registers. This must be done even for call return instructions |
| // since those are not handled by the special code above. |
| if (instrNeedsSpills) |
| for (unsigned OpNum=0; OpNum < MInst->getNumOperands(); ++OpNum) { |
| MachineOperand& Op = MInst->getOperand(OpNum); |
| if (Op.getType() == MachineOperand::MO_VirtualRegister || |
| Op.getType() == MachineOperand::MO_CCRegister) { |
| const Value* Val = Op.getVRegValue(); |
| if (const LiveRange *LR = LRI->getLiveRangeForValue(Val)) |
| if (LR->isMarkedForSpill()) |
| insertCode4SpilledLR(LR, MII, MBB, OpNum); |
| } |
| } // for each operand |
| } |
| |
| /// Iterate over all the MachineBasicBlocks in the current function and set |
| /// the allocated registers for each instruction (using updateInstruction()), |
| /// after register allocation is complete. Then move code out of delay slots. |
| /// |
| void PhyRegAlloc::updateMachineCode() |
| { |
| // Insert any instructions needed at method entry |
| MachineBasicBlock::iterator MII = MF->front().begin(); |
| PrependInstructions(AddedInstrAtEntry.InstrnsBefore, MF->front(), MII, |
| "At function entry: \n"); |
| assert(AddedInstrAtEntry.InstrnsAfter.empty() && |
| "InstrsAfter should be unnecessary since we are just inserting at " |
| "the function entry point here."); |
| |
| for (MachineFunction::iterator BBI = MF->begin(), BBE = MF->end(); |
| BBI != BBE; ++BBI) { |
| MachineBasicBlock &MBB = *BBI; |
| |
| // Iterate over all machine instructions in BB and mark operands with |
| // their assigned registers or insert spill code, as appropriate. |
| // Also, fix operands of call/return instructions. |
| for (MachineBasicBlock::iterator MII = MBB.begin(); MII != MBB.end(); ++MII) |
| if (! TM.getInstrInfo().isDummyPhiInstr((*MII)->getOpCode())) |
| updateInstruction(MII, MBB); |
| |
| // Now, move code out of delay slots of branches and returns if needed. |
| // (Also, move "after" code from calls to the last delay slot instruction.) |
| // Moving code out of delay slots is needed in 2 situations: |
| // (1) If this is a branch and it needs instructions inserted after it, |
| // move any existing instructions out of the delay slot so that the |
| // instructions can go into the delay slot. This only supports the |
| // case that #instrsAfter <= #delay slots. |
| // |
| // (2) If any instruction in the delay slot needs |
| // instructions inserted, move it out of the delay slot and before the |
| // branch because putting code before or after it would be VERY BAD! |
| // |
| // If the annul bit of the branch is set, neither of these is legal! |
| // If so, we need to handle spill differently but annulling is not yet used. |
| for (MachineBasicBlock::iterator MII = MBB.begin(); |
| MII != MBB.end(); ++MII) |
| if (unsigned delaySlots = |
| TM.getInstrInfo().getNumDelaySlots((*MII)->getOpCode())) { |
| MachineInstr *MInst = *MII, *DelaySlotMI = *(MII+1); |
| |
| // Check the 2 conditions above: |
| // (1) Does a branch need instructions added after it? |
| // (2) O/w does delay slot instr. need instrns before or after? |
| bool isBranch = (TM.getInstrInfo().isBranch(MInst->getOpCode()) || |
| TM.getInstrInfo().isReturn(MInst->getOpCode())); |
| bool cond1 = (isBranch && |
| AddedInstrMap.count(MInst) && |
| AddedInstrMap[MInst].InstrnsAfter.size() > 0); |
| bool cond2 = (AddedInstrMap.count(DelaySlotMI) && |
| (AddedInstrMap[DelaySlotMI].InstrnsBefore.size() > 0 || |
| AddedInstrMap[DelaySlotMI].InstrnsAfter.size() > 0)); |
| |
| if (cond1 || cond2) { |
| assert((MInst->getOpCodeFlags() & AnnulFlag) == 0 && |
| "FIXME: Moving an annulled delay slot instruction!"); |
| assert(delaySlots==1 && |
| "InsertBefore does not yet handle >1 delay slots!"); |
| InsertBefore(DelaySlotMI, MBB, MII); // MII pts back to branch |
| |
| // In case (1), delete it and don't replace with anything! |
| // Otherwise (i.e., case (2) only) replace it with a NOP. |
| if (cond1) { |
| DeleteInstruction(MBB, ++MII); // MII now points to next inst. |
| --MII; // reset MII for ++MII of loop |
| } |
| else |
| SubstituteInPlace(BuildMI(TM.getInstrInfo().getNOPOpCode(),1), |
| MBB, MII+1); // replace with NOP |
| |
| if (DEBUG_RA) { |
| std::cerr << "\nRegAlloc: Moved instr. with added code: " |
| << *DelaySlotMI |
| << " out of delay slots of instr: " << *MInst; |
| } |
| } |
| else |
| // For non-branch instr with delay slots (probably a call), move |
| // InstrAfter to the instr. in the last delay slot. |
| move2DelayedInstr(*MII, *(MII+delaySlots)); |
| } |
| |
| // Finally iterate over all instructions in BB and insert before/after |
| for (MachineBasicBlock::iterator MII=MBB.begin(); MII != MBB.end(); ++MII) { |
| MachineInstr *MInst = *MII; |
| |
| // do not process Phis |
| if (TM.getInstrInfo().isDummyPhiInstr(MInst->getOpCode())) |
| continue; |
| |
| // if there are any added instructions... |
| if (AddedInstrMap.count(MInst)) { |
| AddedInstrns &CallAI = AddedInstrMap[MInst]; |
| |
| #ifndef NDEBUG |
| bool isBranch = (TM.getInstrInfo().isBranch(MInst->getOpCode()) || |
| TM.getInstrInfo().isReturn(MInst->getOpCode())); |
| assert((!isBranch || |
| AddedInstrMap[MInst].InstrnsAfter.size() <= |
| TM.getInstrInfo().getNumDelaySlots(MInst->getOpCode())) && |
| "Cannot put more than #delaySlots instrns after " |
| "branch or return! Need to handle temps differently."); |
| #endif |
| |
| #ifndef NDEBUG |
| // Temporary sanity checking code to detect whether the same machine |
| // instruction is ever inserted twice before/after a call. |
| // I suspect this is happening but am not sure. --Vikram, 7/1/03. |
| std::set<const MachineInstr*> instrsSeen; |
| for (int i = 0, N = CallAI.InstrnsBefore.size(); i < N; ++i) { |
| assert(instrsSeen.count(CallAI.InstrnsBefore[i]) == 0 && |
| "Duplicate machine instruction in InstrnsBefore!"); |
| instrsSeen.insert(CallAI.InstrnsBefore[i]); |
| } |
| for (int i = 0, N = CallAI.InstrnsAfter.size(); i < N; ++i) { |
| assert(instrsSeen.count(CallAI.InstrnsAfter[i]) == 0 && |
| "Duplicate machine instruction in InstrnsBefore/After!"); |
| instrsSeen.insert(CallAI.InstrnsAfter[i]); |
| } |
| #endif |
| |
| // Now add the instructions before/after this MI. |
| // We do this here to ensure that spill for an instruction is inserted |
| // as close as possible to an instruction (see above insertCode4Spill) |
| if (! CallAI.InstrnsBefore.empty()) |
| PrependInstructions(CallAI.InstrnsBefore, MBB, MII,""); |
| |
| if (! CallAI.InstrnsAfter.empty()) |
| AppendInstructions(CallAI.InstrnsAfter, MBB, MII,""); |
| |
| } // if there are any added instructions |
| } // for each machine instruction |
| } |
| } |
| |
| |
| /// Insert spill code for AN operand whose LR was spilled. May be called |
| /// repeatedly for a single MachineInstr if it has many spilled operands. On |
| /// each call, it finds a register which is not live at that instruction and |
| /// also which is not used by other spilled operands of the same |
| /// instruction. Then it uses this register temporarily to accommodate the |
| /// spilled value. |
| /// |
| void PhyRegAlloc::insertCode4SpilledLR(const LiveRange *LR, |
| MachineBasicBlock::iterator& MII, |
| MachineBasicBlock &MBB, |
| const unsigned OpNum) { |
| MachineInstr *MInst = *MII; |
| const BasicBlock *BB = MBB.getBasicBlock(); |
| |
| assert((! TM.getInstrInfo().isCall(MInst->getOpCode()) || OpNum == 0) && |
| "Outgoing arg of a call must be handled elsewhere (func arg ok)"); |
| assert(! TM.getInstrInfo().isReturn(MInst->getOpCode()) && |
| "Return value of a ret must be handled elsewhere"); |
| |
| MachineOperand& Op = MInst->getOperand(OpNum); |
| bool isDef = Op.opIsDefOnly(); |
| bool isDefAndUse = Op.opIsDefAndUse(); |
| unsigned RegType = MRI.getRegTypeForLR(LR); |
| int SpillOff = LR->getSpillOffFromFP(); |
| RegClass *RC = LR->getRegClass(); |
| |
| // Get the live-variable set to find registers free before this instr. |
| const ValueSet &LVSetBef = LVI->getLiveVarSetBeforeMInst(MInst, BB); |
| |
| #ifndef NDEBUG |
| // If this instr. is in the delay slot of a branch or return, we need to |
| // include all live variables before that branch or return -- we don't want to |
| // trample those! Verify that the set is included in the LV set before MInst. |
| if (MII != MBB.begin()) { |
| MachineInstr *PredMI = *(MII-1); |
| if (unsigned DS = TM.getInstrInfo().getNumDelaySlots(PredMI->getOpCode())) |
| assert(set_difference(LVI->getLiveVarSetBeforeMInst(PredMI), LVSetBef) |
| .empty() && "Live-var set before branch should be included in " |
| "live-var set of each delay slot instruction!"); |
| } |
| #endif |
| |
| MF->getInfo()->pushTempValue(MRI.getSpilledRegSize(RegType)); |
| |
| std::vector<MachineInstr*> MIBef, MIAft; |
| std::vector<MachineInstr*> AdIMid; |
| |
| // Choose a register to hold the spilled value, if one was not preallocated. |
| // This may insert code before and after MInst to free up the value. If so, |
| // this code should be first/last in the spill sequence before/after MInst. |
| int TmpRegU=(LR->hasColor() |
| ? MRI.getUnifiedRegNum(LR->getRegClassID(),LR->getColor()) |
| : getUsableUniRegAtMI(RegType, &LVSetBef, MInst, MIBef,MIAft)); |
| |
| // Set the operand first so that it this register does not get used |
| // as a scratch register for later calls to getUsableUniRegAtMI below |
| MInst->SetRegForOperand(OpNum, TmpRegU); |
| |
| // get the added instructions for this instruction |
| AddedInstrns &AI = AddedInstrMap[MInst]; |
| |
| // We may need a scratch register to copy the spilled value to/from memory. |
| // This may itself have to insert code to free up a scratch register. |
| // Any such code should go before (after) the spill code for a load (store). |
| // The scratch reg is not marked as used because it is only used |
| // for the copy and not used across MInst. |
| int scratchRegType = -1; |
| int scratchReg = -1; |
| if (MRI.regTypeNeedsScratchReg(RegType, scratchRegType)) { |
| scratchReg = getUsableUniRegAtMI(scratchRegType, &LVSetBef, |
| MInst, MIBef, MIAft); |
| assert(scratchReg != MRI.getInvalidRegNum()); |
| } |
| |
| if (!isDef || isDefAndUse) { |
| // for a USE, we have to load the value of LR from stack to a TmpReg |
| // and use the TmpReg as one operand of instruction |
| |
| // actual loading instruction(s) |
| MRI.cpMem2RegMI(AdIMid, MRI.getFramePointer(), SpillOff, TmpRegU, |
| RegType, scratchReg); |
| |
| // the actual load should be after the instructions to free up TmpRegU |
| MIBef.insert(MIBef.end(), AdIMid.begin(), AdIMid.end()); |
| AdIMid.clear(); |
| } |
| |
| if (isDef || isDefAndUse) { // if this is a Def |
| // for a DEF, we have to store the value produced by this instruction |
| // on the stack position allocated for this LR |
| |
| // actual storing instruction(s) |
| MRI.cpReg2MemMI(AdIMid, TmpRegU, MRI.getFramePointer(), SpillOff, |
| RegType, scratchReg); |
| |
| MIAft.insert(MIAft.begin(), AdIMid.begin(), AdIMid.end()); |
| } // if !DEF |
| |
| // Finally, insert the entire spill code sequences before/after MInst |
| AI.InstrnsBefore.insert(AI.InstrnsBefore.end(), MIBef.begin(), MIBef.end()); |
| AI.InstrnsAfter.insert(AI.InstrnsAfter.begin(), MIAft.begin(), MIAft.end()); |
| |
| if (DEBUG_RA) { |
| std::cerr << "\nFor Inst:\n " << *MInst; |
| std::cerr << "SPILLED LR# " << LR->getUserIGNode()->getIndex(); |
| std::cerr << "; added Instructions:"; |
| for_each(MIBef.begin(), MIBef.end(), std::mem_fun(&MachineInstr::dump)); |
| for_each(MIAft.begin(), MIAft.end(), std::mem_fun(&MachineInstr::dump)); |
| } |
| } |
| |
| |
| /// Insert caller saving/restoring instructions before/after a call machine |
| /// instruction (before or after any other instructions that were inserted for |
| /// the call). |
| /// |
| void |
| PhyRegAlloc::insertCallerSavingCode(std::vector<MachineInstr*> &instrnsBefore, |
| std::vector<MachineInstr*> &instrnsAfter, |
| MachineInstr *CallMI, |
| const BasicBlock *BB) { |
| assert(TM.getInstrInfo().isCall(CallMI->getOpCode())); |
| |
| // hash set to record which registers were saved/restored |
| hash_set<unsigned> PushedRegSet; |
| |
| CallArgsDescriptor* argDesc = CallArgsDescriptor::get(CallMI); |
| |
| // if the call is to a instrumentation function, do not insert save and |
| // restore instructions the instrumentation function takes care of save |
| // restore for volatile regs. |
| // |
| // FIXME: this should be made general, not specific to the reoptimizer! |
| const Function *Callee = argDesc->getCallInst()->getCalledFunction(); |
| bool isLLVMFirstTrigger = Callee && Callee->getName() == "llvm_first_trigger"; |
| |
| // Now check if the call has a return value (using argDesc) and if so, |
| // find the LR of the TmpInstruction representing the return value register. |
| // (using the last or second-last *implicit operand* of the call MI). |
| // Insert it to to the PushedRegSet since we must not save that register |
| // and restore it after the call. |
| // We do this because, we look at the LV set *after* the instruction |
| // to determine, which LRs must be saved across calls. The return value |
| // of the call is live in this set - but we must not save/restore it. |
| if (const Value *origRetVal = argDesc->getReturnValue()) { |
| unsigned retValRefNum = (CallMI->getNumImplicitRefs() - |
| (argDesc->getIndirectFuncPtr()? 1 : 2)); |
| const TmpInstruction* tmpRetVal = |
| cast<TmpInstruction>(CallMI->getImplicitRef(retValRefNum)); |
| assert(tmpRetVal->getOperand(0) == origRetVal && |
| tmpRetVal->getType() == origRetVal->getType() && |
| "Wrong implicit ref?"); |
| LiveRange *RetValLR = LRI->getLiveRangeForValue(tmpRetVal); |
| assert(RetValLR && "No LR for RetValue of call"); |
| |
| if (! RetValLR->isMarkedForSpill()) |
| PushedRegSet.insert(MRI.getUnifiedRegNum(RetValLR->getRegClassID(), |
| RetValLR->getColor())); |
| } |
| |
| const ValueSet &LVSetAft = LVI->getLiveVarSetAfterMInst(CallMI, BB); |
| ValueSet::const_iterator LIt = LVSetAft.begin(); |
| |
| // for each live var in live variable set after machine inst |
| for( ; LIt != LVSetAft.end(); ++LIt) { |
| // get the live range corresponding to live var |
| LiveRange *const LR = LRI->getLiveRangeForValue(*LIt); |
| |
| // LR can be null if it is a const since a const |
| // doesn't have a dominating def - see Assumptions above |
| if (LR) { |
| if (! LR->isMarkedForSpill()) { |
| assert(LR->hasColor() && "LR is neither spilled nor colored?"); |
| unsigned RCID = LR->getRegClassID(); |
| unsigned Color = LR->getColor(); |
| |
| if (MRI.isRegVolatile(RCID, Color) ) { |
| // if this is a call to the first-level reoptimizer |
| // instrumentation entry point, and the register is not |
| // modified by call, don't save and restore it. |
| if (isLLVMFirstTrigger && !MRI.modifiedByCall(RCID, Color)) |
| continue; |
| |
| // if the value is in both LV sets (i.e., live before and after |
| // the call machine instruction) |
| unsigned Reg = MRI.getUnifiedRegNum(RCID, Color); |
| |
| // if we haven't already pushed this register... |
| if( PushedRegSet.find(Reg) == PushedRegSet.end() ) { |
| unsigned RegType = MRI.getRegTypeForLR(LR); |
| |
| // Now get two instructions - to push on stack and pop from stack |
| // and add them to InstrnsBefore and InstrnsAfter of the |
| // call instruction |
| int StackOff = |
| MF->getInfo()->pushTempValue(MRI.getSpilledRegSize(RegType)); |
| |
| //---- Insert code for pushing the reg on stack ---------- |
| |
| std::vector<MachineInstr*> AdIBef, AdIAft; |
| |
| // We may need a scratch register to copy the saved value |
| // to/from memory. This may itself have to insert code to |
| // free up a scratch register. Any such code should go before |
| // the save code. The scratch register, if any, is by default |
| // temporary and not "used" by the instruction unless the |
| // copy code itself decides to keep the value in the scratch reg. |
| int scratchRegType = -1; |
| int scratchReg = -1; |
| if (MRI.regTypeNeedsScratchReg(RegType, scratchRegType)) |
| { // Find a register not live in the LVSet before CallMI |
| const ValueSet &LVSetBef = |
| LVI->getLiveVarSetBeforeMInst(CallMI, BB); |
| scratchReg = getUsableUniRegAtMI(scratchRegType, &LVSetBef, |
| CallMI, AdIBef, AdIAft); |
| assert(scratchReg != MRI.getInvalidRegNum()); |
| } |
| |
| if (AdIBef.size() > 0) |
| instrnsBefore.insert(instrnsBefore.end(), |
| AdIBef.begin(), AdIBef.end()); |
| |
| MRI.cpReg2MemMI(instrnsBefore, Reg, MRI.getFramePointer(), |
| StackOff, RegType, scratchReg); |
| |
| if (AdIAft.size() > 0) |
| instrnsBefore.insert(instrnsBefore.end(), |
| AdIAft.begin(), AdIAft.end()); |
| |
| //---- Insert code for popping the reg from the stack ---------- |
| AdIBef.clear(); |
| AdIAft.clear(); |
| |
| // We may need a scratch register to copy the saved value |
| // from memory. This may itself have to insert code to |
| // free up a scratch register. Any such code should go |
| // after the save code. As above, scratch is not marked "used". |
| scratchRegType = -1; |
| scratchReg = -1; |
| if (MRI.regTypeNeedsScratchReg(RegType, scratchRegType)) |
| { // Find a register not live in the LVSet after CallMI |
| scratchReg = getUsableUniRegAtMI(scratchRegType, &LVSetAft, |
| CallMI, AdIBef, AdIAft); |
| assert(scratchReg != MRI.getInvalidRegNum()); |
| } |
| |
| if (AdIBef.size() > 0) |
| instrnsAfter.insert(instrnsAfter.end(), |
| AdIBef.begin(), AdIBef.end()); |
| |
| MRI.cpMem2RegMI(instrnsAfter, MRI.getFramePointer(), StackOff, |
| Reg, RegType, scratchReg); |
| |
| if (AdIAft.size() > 0) |
| instrnsAfter.insert(instrnsAfter.end(), |
| AdIAft.begin(), AdIAft.end()); |
| |
| PushedRegSet.insert(Reg); |
| |
| if(DEBUG_RA) { |
| std::cerr << "\nFor call inst:" << *CallMI; |
| std::cerr << " -inserted caller saving instrs: Before:\n\t "; |
| for_each(instrnsBefore.begin(), instrnsBefore.end(), |
| std::mem_fun(&MachineInstr::dump)); |
| std::cerr << " -and After:\n\t "; |
| for_each(instrnsAfter.begin(), instrnsAfter.end(), |
| std::mem_fun(&MachineInstr::dump)); |
| } |
| } // if not already pushed |
| } // if LR has a volatile color |
| } // if LR has color |
| } // if there is a LR for Var |
| } // for each value in the LV set after instruction |
| } |
| |
| |
| /// Returns the unified register number of a temporary register to be used |
| /// BEFORE MInst. If no register is available, it will pick one and modify |
| /// MIBef and MIAft to contain instructions used to free up this returned |
| /// register. |
| /// |
| int PhyRegAlloc::getUsableUniRegAtMI(const int RegType, |
| const ValueSet *LVSetBef, |
| MachineInstr *MInst, |
| std::vector<MachineInstr*>& MIBef, |
| std::vector<MachineInstr*>& MIAft) { |
| RegClass* RC = getRegClassByID(MRI.getRegClassIDOfRegType(RegType)); |
| |
| int RegU = getUnusedUniRegAtMI(RC, RegType, MInst, LVSetBef); |
| |
| if (RegU == -1) { |
| // we couldn't find an unused register. Generate code to free up a reg by |
| // saving it on stack and restoring after the instruction |
| |
| int TmpOff = MF->getInfo()->pushTempValue(MRI.getSpilledRegSize(RegType)); |
| |
| RegU = getUniRegNotUsedByThisInst(RC, RegType, MInst); |
| |
| // Check if we need a scratch register to copy this register to memory. |
| int scratchRegType = -1; |
| if (MRI.regTypeNeedsScratchReg(RegType, scratchRegType)) { |
| int scratchReg = getUsableUniRegAtMI(scratchRegType, LVSetBef, |
| MInst, MIBef, MIAft); |
| assert(scratchReg != MRI.getInvalidRegNum()); |
| |
| // We may as well hold the value in the scratch register instead |
| // of copying it to memory and back. But we have to mark the |
| // register as used by this instruction, so it does not get used |
| // as a scratch reg. by another operand or anyone else. |
| ScratchRegsUsed.insert(std::make_pair(MInst, scratchReg)); |
| MRI.cpReg2RegMI(MIBef, RegU, scratchReg, RegType); |
| MRI.cpReg2RegMI(MIAft, scratchReg, RegU, RegType); |
| } else { // the register can be copied directly to/from memory so do it. |
| MRI.cpReg2MemMI(MIBef, RegU, MRI.getFramePointer(), TmpOff, RegType); |
| MRI.cpMem2RegMI(MIAft, MRI.getFramePointer(), TmpOff, RegU, RegType); |
| } |
| } |
| |
| return RegU; |
| } |
| |
| |
| /// Returns the register-class register number of a new unused register that |
| /// can be used to accommodate a temporary value. May be called repeatedly |
| /// for a single MachineInstr. On each call, it finds a register which is not |
| /// live at that instruction and which is not used by any spilled operands of |
| /// that instruction. |
| /// |
| int PhyRegAlloc::getUnusedUniRegAtMI(RegClass *RC, const int RegType, |
| const MachineInstr *MInst, |
| const ValueSet* LVSetBef) { |
| RC->clearColorsUsed(); // Reset array |
| |
| if (LVSetBef == NULL) { |
| LVSetBef = &LVI->getLiveVarSetBeforeMInst(MInst); |
| assert(LVSetBef != NULL && "Unable to get live-var set before MInst?"); |
| } |
| |
| ValueSet::const_iterator LIt = LVSetBef->begin(); |
| |
| // for each live var in live variable set after machine inst |
| for ( ; LIt != LVSetBef->end(); ++LIt) { |
| // Get the live range corresponding to live var, and its RegClass |
| LiveRange *const LRofLV = LRI->getLiveRangeForValue(*LIt ); |
| |
| // LR can be null if it is a const since a const |
| // doesn't have a dominating def - see Assumptions above |
| if (LRofLV && LRofLV->getRegClass() == RC && LRofLV->hasColor()) |
| RC->markColorsUsed(LRofLV->getColor(), |
| MRI.getRegTypeForLR(LRofLV), RegType); |
| } |
| |
| // It is possible that one operand of this MInst was already spilled |
| // and it received some register temporarily. If that's the case, |
| // it is recorded in machine operand. We must skip such registers. |
| setRelRegsUsedByThisInst(RC, RegType, MInst); |
| |
| int unusedReg = RC->getUnusedColor(RegType); // find first unused color |
| if (unusedReg >= 0) |
| return MRI.getUnifiedRegNum(RC->getID(), unusedReg); |
| |
| return -1; |
| } |
| |
| |
| /// Return the unified register number of a register in class RC which is not |
| /// used by any operands of MInst. |
| /// |
| int PhyRegAlloc::getUniRegNotUsedByThisInst(RegClass *RC, |
| const int RegType, |
| const MachineInstr *MInst) { |
| RC->clearColorsUsed(); |
| |
| setRelRegsUsedByThisInst(RC, RegType, MInst); |
| |
| // find the first unused color |
| int unusedReg = RC->getUnusedColor(RegType); |
| assert(unusedReg >= 0 && |
| "FATAL: No free register could be found in reg class!!"); |
| |
| return MRI.getUnifiedRegNum(RC->getID(), unusedReg); |
| } |
| |
| |
| /// Modify the IsColorUsedArr of register class RC, by setting the bits |
| /// corresponding to register RegNo. This is a helper method of |
| /// setRelRegsUsedByThisInst(). |
| /// |
| static void markRegisterUsed(int RegNo, RegClass *RC, int RegType, |
| const TargetRegInfo &TRI) { |
| unsigned classId = 0; |
| int classRegNum = TRI.getClassRegNum(RegNo, classId); |
| if (RC->getID() == classId) |
| RC->markColorsUsed(classRegNum, RegType, RegType); |
| } |
| |
| void PhyRegAlloc::setRelRegsUsedByThisInst(RegClass *RC, int RegType, |
| const MachineInstr *MI) { |
| assert(OperandsColoredMap[MI] == true && |
| "Illegal to call setRelRegsUsedByThisInst() until colored operands " |
| "are marked for an instruction."); |
| |
| // Add the registers already marked as used by the instruction. Both |
| // explicit and implicit operands are set. |
| for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) |
| if (MI->getOperand(i).hasAllocatedReg()) |
| markRegisterUsed(MI->getOperand(i).getAllocatedRegNum(), RC, RegType,MRI); |
| |
| for (unsigned i = 0, e = MI->getNumImplicitRefs(); i != e; ++i) |
| if (MI->getImplicitOp(i).hasAllocatedReg()) |
| markRegisterUsed(MI->getImplicitOp(i).getAllocatedRegNum(), RC, |
| RegType,MRI); |
| |
| // Add all of the scratch registers that are used to save values across the |
| // instruction (e.g., for saving state register values). |
| std::pair<ScratchRegsUsedTy::iterator, ScratchRegsUsedTy::iterator> |
| IR = ScratchRegsUsed.equal_range(MI); |
| for (ScratchRegsUsedTy::iterator I = IR.first; I != IR.second; ++I) |
| markRegisterUsed(I->second, RC, RegType, MRI); |
| |
| // If there are implicit references, mark their allocated regs as well |
| for (unsigned z=0; z < MI->getNumImplicitRefs(); z++) |
| if (const LiveRange* |
| LRofImpRef = LRI->getLiveRangeForValue(MI->getImplicitRef(z))) |
| if (LRofImpRef->hasColor()) |
| // this implicit reference is in a LR that received a color |
| RC->markColorsUsed(LRofImpRef->getColor(), |
| MRI.getRegTypeForLR(LRofImpRef), RegType); |
| } |
| |
| |
| /// If there are delay slots for an instruction, the instructions added after |
| /// it must really go after the delayed instruction(s). So, we Move the |
| /// InstrAfter of that instruction to the corresponding delayed instruction |
| /// using the following method. |
| /// |
| void PhyRegAlloc::move2DelayedInstr(const MachineInstr *OrigMI, |
| const MachineInstr *DelayedMI) |
| { |
| // "added after" instructions of the original instr |
| std::vector<MachineInstr *> &OrigAft = AddedInstrMap[OrigMI].InstrnsAfter; |
| |
| if (DEBUG_RA && OrigAft.size() > 0) { |
| std::cerr << "\nRegAlloc: Moved InstrnsAfter for: " << *OrigMI; |
| std::cerr << " to last delay slot instrn: " << *DelayedMI; |
| } |
| |
| // "added after" instructions of the delayed instr |
| std::vector<MachineInstr *> &DelayedAft=AddedInstrMap[DelayedMI].InstrnsAfter; |
| |
| // go thru all the "added after instructions" of the original instruction |
| // and append them to the "added after instructions" of the delayed |
| // instructions |
| DelayedAft.insert(DelayedAft.end(), OrigAft.begin(), OrigAft.end()); |
| |
| // empty the "added after instructions" of the original instruction |
| OrigAft.clear(); |
| } |
| |
| |
| void PhyRegAlloc::colorIncomingArgs() |
| { |
| MRI.colorMethodArgs(Fn, *LRI, AddedInstrAtEntry.InstrnsBefore, |
| AddedInstrAtEntry.InstrnsAfter); |
| } |
| |
| |
| /// Determine whether the suggested color of each live range is really usable, |
| /// and then call its setSuggestedColorUsable() method to record the answer. A |
| /// suggested color is NOT usable when the suggested color is volatile AND |
| /// when there are call interferences. |
| /// |
| void PhyRegAlloc::markUnusableSugColors() |
| { |
| LiveRangeMapType::const_iterator HMI = (LRI->getLiveRangeMap())->begin(); |
| LiveRangeMapType::const_iterator HMIEnd = (LRI->getLiveRangeMap())->end(); |
| |
| for (; HMI != HMIEnd ; ++HMI ) { |
| if (HMI->first) { |
| LiveRange *L = HMI->second; // get the LiveRange |
| if (L && L->hasSuggestedColor ()) |
| L->setSuggestedColorUsable |
| (!(MRI.isRegVolatile (L->getRegClassID (), L->getSuggestedColor ()) |
| && L->isCallInterference ())); |
| } |
| } // for all LR's in hash map |
| } |
| |
| |
| /// For each live range that is spilled, allocates a new spill position on the |
| /// stack, and set the stack offsets of the live range that will be spilled to |
| /// that position. This must be called just after coloring the LRs. |
| /// |
| void PhyRegAlloc::allocateStackSpace4SpilledLRs() { |
| if (DEBUG_RA) std::cerr << "\nSetting LR stack offsets for spills...\n"; |
| |
| LiveRangeMapType::const_iterator HMI = LRI->getLiveRangeMap()->begin(); |
| LiveRangeMapType::const_iterator HMIEnd = LRI->getLiveRangeMap()->end(); |
| |
| for ( ; HMI != HMIEnd ; ++HMI) { |
| if (HMI->first && HMI->second) { |
| LiveRange *L = HMI->second; // get the LiveRange |
| if (L->isMarkedForSpill()) { // NOTE: allocating size of long Type ** |
| int stackOffset = MF->getInfo()->allocateSpilledValue(Type::LongTy); |
| L->setSpillOffFromFP(stackOffset); |
| if (DEBUG_RA) |
| std::cerr << " LR# " << L->getUserIGNode()->getIndex() |
| << ": stack-offset = " << stackOffset << "\n"; |
| } |
| } |
| } // for all LR's in hash map |
| } |
| |
| |
| void PhyRegAlloc::saveStateForValue (std::vector<AllocInfo> &state, |
| const Value *V, unsigned Insn, int Opnd) { |
| LiveRangeMapType::const_iterator HMI = LRI->getLiveRangeMap ()->find (V); |
| LiveRangeMapType::const_iterator HMIEnd = LRI->getLiveRangeMap ()->end (); |
| AllocInfo::AllocStateTy AllocState = AllocInfo::NotAllocated; |
| int Placement = -1; |
| if ((HMI != HMIEnd) && HMI->second) { |
| LiveRange *L = HMI->second; |
| assert ((L->hasColor () || L->isMarkedForSpill ()) |
| && "Live range exists but not colored or spilled"); |
| if (L->hasColor ()) { |
| AllocState = AllocInfo::Allocated; |
| Placement = MRI.getUnifiedRegNum (L->getRegClassID (), |
| L->getColor ()); |
| } else if (L->isMarkedForSpill ()) { |
| AllocState = AllocInfo::Spilled; |
| assert (L->hasSpillOffset () |
| && "Live range marked for spill but has no spill offset"); |
| Placement = L->getSpillOffFromFP (); |
| } |
| } |
| state.push_back (AllocInfo (Insn, Opnd, AllocState, Placement)); |
| } |
| |
| |
| /// Save the global register allocation decisions made by the register |
| /// allocator so that they can be accessed later (sort of like "poor man's |
| /// debug info"). |
| /// |
| void PhyRegAlloc::saveState () { |
| std::vector<AllocInfo> &state = FnAllocState[Fn]; |
| unsigned Insn = 0; |
| for (const_inst_iterator II=inst_begin (Fn), IE=inst_end (Fn); II!=IE; ++II){ |
| saveStateForValue (state, (*II), Insn, -1); |
| for (unsigned i = 0; i < (*II)->getNumOperands (); ++i) { |
| const Value *V = (*II)->getOperand (i); |
| // Don't worry about it unless it's something whose reg. we'll need. |
| if (!isa<Argument> (V) && !isa<Instruction> (V)) |
| continue; |
| saveStateForValue (state, V, Insn, i); |
| } |
| ++Insn; |
| } |
| } |
| |
| |
| /// Check the saved state filled in by saveState(), and abort if it looks |
| /// wrong. Only used when debugging. FIXME: Currently it just prints out |
| /// the state, which isn't quite as useful. |
| /// |
| void PhyRegAlloc::verifySavedState () { |
| std::vector<AllocInfo> &state = FnAllocState[Fn]; |
| unsigned Insn = 0; |
| for (const_inst_iterator II=inst_begin (Fn), IE=inst_end (Fn); II!=IE; ++II) { |
| const Instruction *I = *II; |
| MachineCodeForInstruction &Instrs = MachineCodeForInstruction::get (I); |
| std::cerr << "Instruction:\n" << " " << *I << "\n" |
| << "MachineCodeForInstruction:\n"; |
| for (unsigned i = 0, n = Instrs.size (); i != n; ++i) |
| std::cerr << " " << *Instrs[i] << "\n"; |
| std::cerr << "FnAllocState:\n"; |
| for (unsigned i = 0; i < state.size (); ++i) { |
| AllocInfo &S = state[i]; |
| if (Insn == S.Instruction) { |
| std::cerr << " (Instruction " << S.Instruction |
| << ", Operand " << S.Operand |
| << ", AllocState " << S.allocStateToString () |
| << ", Placement " << S.Placement << ")\n"; |
| } |
| } |
| std::cerr << "----------\n"; |
| ++Insn; |
| } |
| } |
| |
| |
| /// Finish the job of saveState(), by collapsing FnAllocState into an LLVM |
| /// Constant and stuffing it inside the Module. (NOTE: Soon, there will be |
| /// other, better ways of storing the saved state; this one is cumbersome and |
| /// does not work well with the JIT.) |
| /// |
| bool PhyRegAlloc::doFinalization (Module &M) { |
| if (!SaveRegAllocState) |
| return false; // Nothing to do here, unless we're saving state. |
| |
| // If saving state into the module, just copy new elements to the |
| // correct global. |
| if (!SaveStateToModule) { |
| ExportedFnAllocState = FnAllocState; |
| // FIXME: should ONLY copy new elements in FnAllocState |
| return false; |
| } |
| |
| // Convert FnAllocState to a single Constant array and add it |
| // to the Module. |
| ArrayType *AT = ArrayType::get (AllocInfo::getConstantType (), 0); |
| std::vector<const Type *> TV; |
| TV.push_back (Type::UIntTy); |
| TV.push_back (AT); |
| PointerType *PT = PointerType::get (StructType::get (TV)); |
| |
| std::vector<Constant *> allstate; |
| for (Module::iterator I = M.begin (), E = M.end (); I != E; ++I) { |
| Function *F = I; |
| if (F->isExternal ()) continue; |
| if (FnAllocState.find (F) == FnAllocState.end ()) { |
| allstate.push_back (ConstantPointerNull::get (PT)); |
| } else { |
| std::vector<AllocInfo> &state = FnAllocState[F]; |
| |
| // Convert state into an LLVM ConstantArray, and put it in a |
| // ConstantStruct (named S) along with its size. |
| std::vector<Constant *> stateConstants; |
| for (unsigned i = 0, s = state.size (); i != s; ++i) |
| stateConstants.push_back (state[i].toConstant ()); |
| unsigned Size = stateConstants.size (); |
| ArrayType *AT = ArrayType::get (AllocInfo::getConstantType (), Size); |
| std::vector<const Type *> TV; |
| TV.push_back (Type::UIntTy); |
| TV.push_back (AT); |
| StructType *ST = StructType::get (TV); |
| std::vector<Constant *> CV; |
| CV.push_back (ConstantUInt::get (Type::UIntTy, Size)); |
| CV.push_back (ConstantArray::get (AT, stateConstants)); |
| Constant *S = ConstantStruct::get (ST, CV); |
| |
| GlobalVariable *GV = |
| new GlobalVariable (ST, true, |
| GlobalValue::InternalLinkage, S, |
| F->getName () + ".regAllocState", &M); |
| |
| // Have: { uint, [Size x { uint, int, uint, int }] } * |
| // Cast it to: { uint, [0 x { uint, int, uint, int }] } * |
| Constant *CE = ConstantExpr::getCast (ConstantPointerRef::get (GV), PT); |
| allstate.push_back (CE); |
| } |
| } |
| |
| unsigned Size = allstate.size (); |
| // Final structure type is: |
| // { uint, [Size x { uint, [0 x { uint, int, uint, int }] } *] } |
| std::vector<const Type *> TV2; |
| TV2.push_back (Type::UIntTy); |
| ArrayType *AT2 = ArrayType::get (PT, Size); |
| TV2.push_back (AT2); |
| StructType *ST2 = StructType::get (TV2); |
| std::vector<Constant *> CV2; |
| CV2.push_back (ConstantUInt::get (Type::UIntTy, Size)); |
| CV2.push_back (ConstantArray::get (AT2, allstate)); |
| new GlobalVariable (ST2, true, GlobalValue::ExternalLinkage, |
| ConstantStruct::get (ST2, CV2), "_llvm_regAllocState", |
| &M); |
| return false; // No error. |
| } |
| |
| |
| /// Allocate registers for the machine code previously generated for F using |
| /// the graph-coloring algorithm. |
| /// |
| bool PhyRegAlloc::runOnFunction (Function &F) { |
| if (DEBUG_RA) |
| std::cerr << "\n********* Function "<< F.getName () << " ***********\n"; |
| |
| Fn = &F; |
| MF = &MachineFunction::get (Fn); |
| LVI = &getAnalysis<FunctionLiveVarInfo> (); |
| LRI = new LiveRangeInfo (Fn, TM, RegClassList); |
| LoopDepthCalc = &getAnalysis<LoopInfo> (); |
| |
| // Create each RegClass for the target machine and add it to the |
| // RegClassList. This must be done before calling constructLiveRanges(). |
| for (unsigned rc = 0; rc != NumOfRegClasses; ++rc) |
| RegClassList.push_back (new RegClass (Fn, &TM.getRegInfo (), |
| MRI.getMachineRegClass (rc))); |
| |
| LRI->constructLiveRanges(); // create LR info |
| if (DEBUG_RA >= RA_DEBUG_LiveRanges) |
| LRI->printLiveRanges(); |
| |
| createIGNodeListsAndIGs(); // create IGNode list and IGs |
| |
| buildInterferenceGraphs(); // build IGs in all reg classes |
| |
| if (DEBUG_RA >= RA_DEBUG_LiveRanges) { |
| // print all LRs in all reg classes |
| for ( unsigned rc=0; rc < NumOfRegClasses ; rc++) |
| RegClassList[rc]->printIGNodeList(); |
| |
| // print IGs in all register classes |
| for ( unsigned rc=0; rc < NumOfRegClasses ; rc++) |
| RegClassList[rc]->printIG(); |
| } |
| |
| LRI->coalesceLRs(); // coalesce all live ranges |
| |
| if (DEBUG_RA >= RA_DEBUG_LiveRanges) { |
| // print all LRs in all reg classes |
| for (unsigned rc=0; rc < NumOfRegClasses; rc++) |
| RegClassList[rc]->printIGNodeList(); |
| |
| // print IGs in all register classes |
| for (unsigned rc=0; rc < NumOfRegClasses; rc++) |
| RegClassList[rc]->printIG(); |
| } |
| |
| // mark un-usable suggested color before graph coloring algorithm. |
| // When this is done, the graph coloring algo will not reserve |
| // suggested color unnecessarily - they can be used by another LR |
| markUnusableSugColors(); |
| |
| // color all register classes using the graph coloring algo |
| for (unsigned rc=0; rc < NumOfRegClasses ; rc++) |
| RegClassList[rc]->colorAllRegs(); |
| |
| // After graph coloring, if some LRs did not receive a color (i.e, spilled) |
| // a position for such spilled LRs |
| allocateStackSpace4SpilledLRs(); |
| |
| // Reset the temp. area on the stack before use by the first instruction. |
| // This will also happen after updating each instruction. |
| MF->getInfo()->popAllTempValues(); |
| |
| // color incoming args - if the correct color was not received |
| // insert code to copy to the correct register |
| colorIncomingArgs(); |
| |
| // Save register allocation state for this function in a Constant. |
| if (SaveRegAllocState) |
| saveState(); |
| if (DEBUG_RA) { // Check our work. |
| verifySavedState (); |
| } |
| |
| // Now update the machine code with register names and add any additional |
| // code inserted by the register allocator to the instruction stream. |
| updateMachineCode(); |
| |
| if (DEBUG_RA) { |
| std::cerr << "\n**** Machine Code After Register Allocation:\n\n"; |
| MF->dump(); |
| } |
| |
| // Tear down temporary data structures |
| for (unsigned rc = 0; rc < NumOfRegClasses; ++rc) |
| delete RegClassList[rc]; |
| RegClassList.clear (); |
| AddedInstrMap.clear (); |
| OperandsColoredMap.clear (); |
| ScratchRegsUsed.clear (); |
| AddedInstrAtEntry.clear (); |
| delete LRI; |
| |
| if (DEBUG_RA) std::cerr << "\nRegister allocation complete!\n"; |
| return false; // Function was not modified |
| } |
| |
| } // End llvm namespace |