//===- HexagonInstrInfo.cpp - Hexagon Instruction Information -------------===// // // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. // See https://llvm.org/LICENSE.txt for license information. // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception // //===----------------------------------------------------------------------===// // // This file contains the Hexagon implementation of the TargetInstrInfo class. // //===----------------------------------------------------------------------===// #include "HexagonInstrInfo.h" #include "Hexagon.h" #include "HexagonFrameLowering.h" #include "HexagonHazardRecognizer.h" #include "HexagonRegisterInfo.h" #include "HexagonSubtarget.h" #include "llvm/ADT/ArrayRef.h" #include "llvm/ADT/SmallPtrSet.h" #include "llvm/ADT/SmallVector.h" #include "llvm/ADT/StringRef.h" #include "llvm/CodeGen/DFAPacketizer.h" #include "llvm/CodeGen/LivePhysRegs.h" #include "llvm/CodeGen/MachineBasicBlock.h" #include "llvm/CodeGen/MachineBranchProbabilityInfo.h" #include "llvm/CodeGen/MachineFrameInfo.h" #include "llvm/CodeGen/MachineFunction.h" #include "llvm/CodeGen/MachineInstr.h" #include "llvm/CodeGen/MachineInstrBuilder.h" #include "llvm/CodeGen/MachineInstrBundle.h" #include "llvm/CodeGen/MachineLoopInfo.h" #include "llvm/CodeGen/MachineMemOperand.h" #include "llvm/CodeGen/MachineOperand.h" #include "llvm/CodeGen/MachineRegisterInfo.h" #include "llvm/CodeGen/ScheduleDAG.h" #include "llvm/CodeGen/TargetInstrInfo.h" #include "llvm/CodeGen/TargetOpcodes.h" #include "llvm/CodeGen/TargetRegisterInfo.h" #include "llvm/CodeGen/TargetSubtargetInfo.h" #include "llvm/IR/DebugLoc.h" #include "llvm/MC/MCAsmInfo.h" #include "llvm/MC/MCInstrDesc.h" #include "llvm/MC/MCInstrItineraries.h" #include "llvm/MC/MCRegisterInfo.h" #include "llvm/Support/BranchProbability.h" #include "llvm/Support/CommandLine.h" #include "llvm/Support/Debug.h" #include "llvm/Support/ErrorHandling.h" #include "llvm/Support/MachineValueType.h" #include "llvm/Support/MathExtras.h" #include "llvm/Support/raw_ostream.h" #include "llvm/Target/TargetMachine.h" #include #include #include #include #include #include #include using namespace llvm; #define DEBUG_TYPE "hexagon-instrinfo" #define GET_INSTRINFO_CTOR_DTOR #define GET_INSTRMAP_INFO #include "HexagonDepTimingClasses.h" #include "HexagonGenDFAPacketizer.inc" #include "HexagonGenInstrInfo.inc" cl::opt ScheduleInlineAsm("hexagon-sched-inline-asm", cl::Hidden, cl::init(false), cl::desc("Do not consider inline-asm a scheduling/" "packetization boundary.")); static cl::opt EnableBranchPrediction("hexagon-enable-branch-prediction", cl::Hidden, cl::init(true), cl::desc("Enable branch prediction")); static cl::opt DisableNVSchedule("disable-hexagon-nv-schedule", cl::Hidden, cl::ZeroOrMore, cl::init(false), cl::desc("Disable schedule adjustment for new value stores.")); static cl::opt EnableTimingClassLatency( "enable-timing-class-latency", cl::Hidden, cl::init(false), cl::desc("Enable timing class latency")); static cl::opt EnableALUForwarding( "enable-alu-forwarding", cl::Hidden, cl::init(true), cl::desc("Enable vec alu forwarding")); static cl::opt EnableACCForwarding( "enable-acc-forwarding", cl::Hidden, cl::init(true), cl::desc("Enable vec acc forwarding")); static cl::opt BranchRelaxAsmLarge("branch-relax-asm-large", cl::init(true), cl::Hidden, cl::ZeroOrMore, cl::desc("branch relax asm")); static cl::opt UseDFAHazardRec("dfa-hazard-rec", cl::init(true), cl::Hidden, cl::ZeroOrMore, cl::desc("Use the DFA based hazard recognizer.")); /// Constants for Hexagon instructions. const int Hexagon_MEMW_OFFSET_MAX = 4095; const int Hexagon_MEMW_OFFSET_MIN = -4096; const int Hexagon_MEMD_OFFSET_MAX = 8191; const int Hexagon_MEMD_OFFSET_MIN = -8192; const int Hexagon_MEMH_OFFSET_MAX = 2047; const int Hexagon_MEMH_OFFSET_MIN = -2048; const int Hexagon_MEMB_OFFSET_MAX = 1023; const int Hexagon_MEMB_OFFSET_MIN = -1024; const int Hexagon_ADDI_OFFSET_MAX = 32767; const int Hexagon_ADDI_OFFSET_MIN = -32768; // Pin the vtable to this file. void HexagonInstrInfo::anchor() {} HexagonInstrInfo::HexagonInstrInfo(HexagonSubtarget &ST) : HexagonGenInstrInfo(Hexagon::ADJCALLSTACKDOWN, Hexagon::ADJCALLSTACKUP), Subtarget(ST) {} namespace llvm { namespace HexagonFUnits { bool isSlot0Only(unsigned units); } } static bool isIntRegForSubInst(unsigned Reg) { return (Reg >= Hexagon::R0 && Reg <= Hexagon::R7) || (Reg >= Hexagon::R16 && Reg <= Hexagon::R23); } static bool isDblRegForSubInst(unsigned Reg, const HexagonRegisterInfo &HRI) { return isIntRegForSubInst(HRI.getSubReg(Reg, Hexagon::isub_lo)) && isIntRegForSubInst(HRI.getSubReg(Reg, Hexagon::isub_hi)); } /// Calculate number of instructions excluding the debug instructions. static unsigned nonDbgMICount(MachineBasicBlock::const_instr_iterator MIB, MachineBasicBlock::const_instr_iterator MIE) { unsigned Count = 0; for (; MIB != MIE; ++MIB) { if (!MIB->isDebugInstr()) ++Count; } return Count; } /// Find the hardware loop instruction used to set-up the specified loop. /// On Hexagon, we have two instructions used to set-up the hardware loop /// (LOOP0, LOOP1) with corresponding endloop (ENDLOOP0, ENDLOOP1) instructions /// to indicate the end of a loop. MachineInstr *HexagonInstrInfo::findLoopInstr(MachineBasicBlock *BB, unsigned EndLoopOp, MachineBasicBlock *TargetBB, SmallPtrSet &Visited) const { unsigned LOOPi; unsigned LOOPr; if (EndLoopOp == Hexagon::ENDLOOP0) { LOOPi = Hexagon::J2_loop0i; LOOPr = Hexagon::J2_loop0r; } else { // EndLoopOp == Hexagon::EndLOOP1 LOOPi = Hexagon::J2_loop1i; LOOPr = Hexagon::J2_loop1r; } // The loop set-up instruction will be in a predecessor block for (MachineBasicBlock *PB : BB->predecessors()) { // If this has been visited, already skip it. if (!Visited.insert(PB).second) continue; if (PB == BB) continue; for (auto I = PB->instr_rbegin(), E = PB->instr_rend(); I != E; ++I) { unsigned Opc = I->getOpcode(); if (Opc == LOOPi || Opc == LOOPr) return &*I; // We've reached a different loop, which means the loop01 has been // removed. if (Opc == EndLoopOp && I->getOperand(0).getMBB() != TargetBB) return nullptr; } // Check the predecessors for the LOOP instruction. if (MachineInstr *Loop = findLoopInstr(PB, EndLoopOp, TargetBB, Visited)) return Loop; } return nullptr; } /// Gather register def/uses from MI. /// This treats possible (predicated) defs as actually happening ones /// (conservatively). static inline void parseOperands(const MachineInstr &MI, SmallVector &Defs, SmallVector &Uses) { Defs.clear(); Uses.clear(); for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) { const MachineOperand &MO = MI.getOperand(i); if (!MO.isReg()) continue; Register Reg = MO.getReg(); if (!Reg) continue; if (MO.isUse()) Uses.push_back(MO.getReg()); if (MO.isDef()) Defs.push_back(MO.getReg()); } } // Position dependent, so check twice for swap. static bool isDuplexPairMatch(unsigned Ga, unsigned Gb) { switch (Ga) { case HexagonII::HSIG_None: default: return false; case HexagonII::HSIG_L1: return (Gb == HexagonII::HSIG_L1 || Gb == HexagonII::HSIG_A); case HexagonII::HSIG_L2: return (Gb == HexagonII::HSIG_L1 || Gb == HexagonII::HSIG_L2 || Gb == HexagonII::HSIG_A); case HexagonII::HSIG_S1: return (Gb == HexagonII::HSIG_L1 || Gb == HexagonII::HSIG_L2 || Gb == HexagonII::HSIG_S1 || Gb == HexagonII::HSIG_A); case HexagonII::HSIG_S2: return (Gb == HexagonII::HSIG_L1 || Gb == HexagonII::HSIG_L2 || Gb == HexagonII::HSIG_S1 || Gb == HexagonII::HSIG_S2 || Gb == HexagonII::HSIG_A); case HexagonII::HSIG_A: return (Gb == HexagonII::HSIG_A); case HexagonII::HSIG_Compound: return (Gb == HexagonII::HSIG_Compound); } return false; } /// isLoadFromStackSlot - If the specified machine instruction is a direct /// load from a stack slot, return the virtual or physical register number of /// the destination along with the FrameIndex of the loaded stack slot. If /// not, return 0. This predicate must return 0 if the instruction has /// any side effects other than loading from the stack slot. unsigned HexagonInstrInfo::isLoadFromStackSlot(const MachineInstr &MI, int &FrameIndex) const { switch (MI.getOpcode()) { default: break; case Hexagon::L2_loadri_io: case Hexagon::L2_loadrd_io: case Hexagon::V6_vL32b_ai: case Hexagon::V6_vL32b_nt_ai: case Hexagon::V6_vL32Ub_ai: case Hexagon::LDriw_pred: case Hexagon::LDriw_ctr: case Hexagon::PS_vloadrq_ai: case Hexagon::PS_vloadrw_ai: case Hexagon::PS_vloadrw_nt_ai: { const MachineOperand OpFI = MI.getOperand(1); if (!OpFI.isFI()) return 0; const MachineOperand OpOff = MI.getOperand(2); if (!OpOff.isImm() || OpOff.getImm() != 0) return 0; FrameIndex = OpFI.getIndex(); return MI.getOperand(0).getReg(); } case Hexagon::L2_ploadrit_io: case Hexagon::L2_ploadrif_io: case Hexagon::L2_ploadrdt_io: case Hexagon::L2_ploadrdf_io: { const MachineOperand OpFI = MI.getOperand(2); if (!OpFI.isFI()) return 0; const MachineOperand OpOff = MI.getOperand(3); if (!OpOff.isImm() || OpOff.getImm() != 0) return 0; FrameIndex = OpFI.getIndex(); return MI.getOperand(0).getReg(); } } return 0; } /// isStoreToStackSlot - If the specified machine instruction is a direct /// store to a stack slot, return the virtual or physical register number of /// the source reg along with the FrameIndex of the loaded stack slot. If /// not, return 0. This predicate must return 0 if the instruction has /// any side effects other than storing to the stack slot. unsigned HexagonInstrInfo::isStoreToStackSlot(const MachineInstr &MI, int &FrameIndex) const { switch (MI.getOpcode()) { default: break; case Hexagon::S2_storerb_io: case Hexagon::S2_storerh_io: case Hexagon::S2_storeri_io: case Hexagon::S2_storerd_io: case Hexagon::V6_vS32b_ai: case Hexagon::V6_vS32Ub_ai: case Hexagon::STriw_pred: case Hexagon::STriw_ctr: case Hexagon::PS_vstorerq_ai: case Hexagon::PS_vstorerw_ai: { const MachineOperand &OpFI = MI.getOperand(0); if (!OpFI.isFI()) return 0; const MachineOperand &OpOff = MI.getOperand(1); if (!OpOff.isImm() || OpOff.getImm() != 0) return 0; FrameIndex = OpFI.getIndex(); return MI.getOperand(2).getReg(); } case Hexagon::S2_pstorerbt_io: case Hexagon::S2_pstorerbf_io: case Hexagon::S2_pstorerht_io: case Hexagon::S2_pstorerhf_io: case Hexagon::S2_pstorerit_io: case Hexagon::S2_pstorerif_io: case Hexagon::S2_pstorerdt_io: case Hexagon::S2_pstorerdf_io: { const MachineOperand &OpFI = MI.getOperand(1); if (!OpFI.isFI()) return 0; const MachineOperand &OpOff = MI.getOperand(2); if (!OpOff.isImm() || OpOff.getImm() != 0) return 0; FrameIndex = OpFI.getIndex(); return MI.getOperand(3).getReg(); } } return 0; } /// This function checks if the instruction or bundle of instructions /// has load from stack slot and returns frameindex and machine memory /// operand of that instruction if true. bool HexagonInstrInfo::hasLoadFromStackSlot( const MachineInstr &MI, SmallVectorImpl &Accesses) const { if (MI.isBundle()) { const MachineBasicBlock *MBB = MI.getParent(); MachineBasicBlock::const_instr_iterator MII = MI.getIterator(); for (++MII; MII != MBB->instr_end() && MII->isInsideBundle(); ++MII) if (TargetInstrInfo::hasLoadFromStackSlot(*MII, Accesses)) return true; return false; } return TargetInstrInfo::hasLoadFromStackSlot(MI, Accesses); } /// This function checks if the instruction or bundle of instructions /// has store to stack slot and returns frameindex and machine memory /// operand of that instruction if true. bool HexagonInstrInfo::hasStoreToStackSlot( const MachineInstr &MI, SmallVectorImpl &Accesses) const { if (MI.isBundle()) { const MachineBasicBlock *MBB = MI.getParent(); MachineBasicBlock::const_instr_iterator MII = MI.getIterator(); for (++MII; MII != MBB->instr_end() && MII->isInsideBundle(); ++MII) if (TargetInstrInfo::hasStoreToStackSlot(*MII, Accesses)) return true; return false; } return TargetInstrInfo::hasStoreToStackSlot(MI, Accesses); } /// This function can analyze one/two way branching only and should (mostly) be /// called by target independent side. /// First entry is always the opcode of the branching instruction, except when /// the Cond vector is supposed to be empty, e.g., when analyzeBranch fails, a /// BB with only unconditional jump. Subsequent entries depend upon the opcode, /// e.g. Jump_c p will have /// Cond[0] = Jump_c /// Cond[1] = p /// HW-loop ENDLOOP: /// Cond[0] = ENDLOOP /// Cond[1] = MBB /// New value jump: /// Cond[0] = Hexagon::CMPEQri_f_Jumpnv_t_V4 -- specific opcode /// Cond[1] = R /// Cond[2] = Imm bool HexagonInstrInfo::analyzeBranch(MachineBasicBlock &MBB, MachineBasicBlock *&TBB, MachineBasicBlock *&FBB, SmallVectorImpl &Cond, bool AllowModify) const { TBB = nullptr; FBB = nullptr; Cond.clear(); // If the block has no terminators, it just falls into the block after it. MachineBasicBlock::instr_iterator I = MBB.instr_end(); if (I == MBB.instr_begin()) return false; // A basic block may looks like this: // // [ insn // EH_LABEL // insn // insn // insn // EH_LABEL // insn ] // // It has two succs but does not have a terminator // Don't know how to handle it. do { --I; if (I->isEHLabel()) // Don't analyze EH branches. return true; } while (I != MBB.instr_begin()); I = MBB.instr_end(); --I; while (I->isDebugInstr()) { if (I == MBB.instr_begin()) return false; --I; } bool JumpToBlock = I->getOpcode() == Hexagon::J2_jump && I->getOperand(0).isMBB(); // Delete the J2_jump if it's equivalent to a fall-through. if (AllowModify && JumpToBlock && MBB.isLayoutSuccessor(I->getOperand(0).getMBB())) { LLVM_DEBUG(dbgs() << "\nErasing the jump to successor block\n";); I->eraseFromParent(); I = MBB.instr_end(); if (I == MBB.instr_begin()) return false; --I; } if (!isUnpredicatedTerminator(*I)) return false; // Get the last instruction in the block. MachineInstr *LastInst = &*I; MachineInstr *SecondLastInst = nullptr; // Find one more terminator if present. while (true) { if (&*I != LastInst && !I->isBundle() && isUnpredicatedTerminator(*I)) { if (!SecondLastInst) SecondLastInst = &*I; else // This is a third branch. return true; } if (I == MBB.instr_begin()) break; --I; } int LastOpcode = LastInst->getOpcode(); int SecLastOpcode = SecondLastInst ? SecondLastInst->getOpcode() : 0; // If the branch target is not a basic block, it could be a tail call. // (It is, if the target is a function.) if (LastOpcode == Hexagon::J2_jump && !LastInst->getOperand(0).isMBB()) return true; if (SecLastOpcode == Hexagon::J2_jump && !SecondLastInst->getOperand(0).isMBB()) return true; bool LastOpcodeHasJMP_c = PredOpcodeHasJMP_c(LastOpcode); bool LastOpcodeHasNVJump = isNewValueJump(*LastInst); if (LastOpcodeHasJMP_c && !LastInst->getOperand(1).isMBB()) return true; // If there is only one terminator instruction, process it. if (LastInst && !SecondLastInst) { if (LastOpcode == Hexagon::J2_jump) { TBB = LastInst->getOperand(0).getMBB(); return false; } if (isEndLoopN(LastOpcode)) { TBB = LastInst->getOperand(0).getMBB(); Cond.push_back(MachineOperand::CreateImm(LastInst->getOpcode())); Cond.push_back(LastInst->getOperand(0)); return false; } if (LastOpcodeHasJMP_c) { TBB = LastInst->getOperand(1).getMBB(); Cond.push_back(MachineOperand::CreateImm(LastInst->getOpcode())); Cond.push_back(LastInst->getOperand(0)); return false; } // Only supporting rr/ri versions of new-value jumps. if (LastOpcodeHasNVJump && (LastInst->getNumExplicitOperands() == 3)) { TBB = LastInst->getOperand(2).getMBB(); Cond.push_back(MachineOperand::CreateImm(LastInst->getOpcode())); Cond.push_back(LastInst->getOperand(0)); Cond.push_back(LastInst->getOperand(1)); return false; } LLVM_DEBUG(dbgs() << "\nCant analyze " << printMBBReference(MBB) << " with one jump\n";); // Otherwise, don't know what this is. return true; } bool SecLastOpcodeHasJMP_c = PredOpcodeHasJMP_c(SecLastOpcode); bool SecLastOpcodeHasNVJump = isNewValueJump(*SecondLastInst); if (SecLastOpcodeHasJMP_c && (LastOpcode == Hexagon::J2_jump)) { if (!SecondLastInst->getOperand(1).isMBB()) return true; TBB = SecondLastInst->getOperand(1).getMBB(); Cond.push_back(MachineOperand::CreateImm(SecondLastInst->getOpcode())); Cond.push_back(SecondLastInst->getOperand(0)); FBB = LastInst->getOperand(0).getMBB(); return false; } // Only supporting rr/ri versions of new-value jumps. if (SecLastOpcodeHasNVJump && (SecondLastInst->getNumExplicitOperands() == 3) && (LastOpcode == Hexagon::J2_jump)) { TBB = SecondLastInst->getOperand(2).getMBB(); Cond.push_back(MachineOperand::CreateImm(SecondLastInst->getOpcode())); Cond.push_back(SecondLastInst->getOperand(0)); Cond.push_back(SecondLastInst->getOperand(1)); FBB = LastInst->getOperand(0).getMBB(); return false; } // If the block ends with two Hexagon:JMPs, handle it. The second one is not // executed, so remove it. if (SecLastOpcode == Hexagon::J2_jump && LastOpcode == Hexagon::J2_jump) { TBB = SecondLastInst->getOperand(0).getMBB(); I = LastInst->getIterator(); if (AllowModify) I->eraseFromParent(); return false; } // If the block ends with an ENDLOOP, and J2_jump, handle it. if (isEndLoopN(SecLastOpcode) && LastOpcode == Hexagon::J2_jump) { TBB = SecondLastInst->getOperand(0).getMBB(); Cond.push_back(MachineOperand::CreateImm(SecondLastInst->getOpcode())); Cond.push_back(SecondLastInst->getOperand(0)); FBB = LastInst->getOperand(0).getMBB(); return false; } LLVM_DEBUG(dbgs() << "\nCant analyze " << printMBBReference(MBB) << " with two jumps";); // Otherwise, can't handle this. return true; } unsigned HexagonInstrInfo::removeBranch(MachineBasicBlock &MBB, int *BytesRemoved) const { assert(!BytesRemoved && "code size not handled"); LLVM_DEBUG(dbgs() << "\nRemoving branches out of " << printMBBReference(MBB)); MachineBasicBlock::iterator I = MBB.end(); unsigned Count = 0; while (I != MBB.begin()) { --I; if (I->isDebugInstr()) continue; // Only removing branches from end of MBB. if (!I->isBranch()) return Count; if (Count && (I->getOpcode() == Hexagon::J2_jump)) llvm_unreachable("Malformed basic block: unconditional branch not last"); MBB.erase(&MBB.back()); I = MBB.end(); ++Count; } return Count; } unsigned HexagonInstrInfo::insertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB, MachineBasicBlock *FBB, ArrayRef Cond, const DebugLoc &DL, int *BytesAdded) const { unsigned BOpc = Hexagon::J2_jump; unsigned BccOpc = Hexagon::J2_jumpt; assert(validateBranchCond(Cond) && "Invalid branching condition"); assert(TBB && "insertBranch must not be told to insert a fallthrough"); assert(!BytesAdded && "code size not handled"); // Check if reverseBranchCondition has asked to reverse this branch // If we want to reverse the branch an odd number of times, we want // J2_jumpf. if (!Cond.empty() && Cond[0].isImm()) BccOpc = Cond[0].getImm(); if (!FBB) { if (Cond.empty()) { // Due to a bug in TailMerging/CFG Optimization, we need to add a // special case handling of a predicated jump followed by an // unconditional jump. If not, Tail Merging and CFG Optimization go // into an infinite loop. MachineBasicBlock *NewTBB, *NewFBB; SmallVector Cond; auto Term = MBB.getFirstTerminator(); if (Term != MBB.end() && isPredicated(*Term) && !analyzeBranch(MBB, NewTBB, NewFBB, Cond, false) && MachineFunction::iterator(NewTBB) == ++MBB.getIterator()) { reverseBranchCondition(Cond); removeBranch(MBB); return insertBranch(MBB, TBB, nullptr, Cond, DL); } BuildMI(&MBB, DL, get(BOpc)).addMBB(TBB); } else if (isEndLoopN(Cond[0].getImm())) { int EndLoopOp = Cond[0].getImm(); assert(Cond[1].isMBB()); // Since we're adding an ENDLOOP, there better be a LOOP instruction. // Check for it, and change the BB target if needed. SmallPtrSet VisitedBBs; MachineInstr *Loop = findLoopInstr(TBB, EndLoopOp, Cond[1].getMBB(), VisitedBBs); assert(Loop != nullptr && "Inserting an ENDLOOP without a LOOP"); Loop->getOperand(0).setMBB(TBB); // Add the ENDLOOP after the finding the LOOP0. BuildMI(&MBB, DL, get(EndLoopOp)).addMBB(TBB); } else if (isNewValueJump(Cond[0].getImm())) { assert((Cond.size() == 3) && "Only supporting rr/ri version of nvjump"); // New value jump // (ins IntRegs:$src1, IntRegs:$src2, brtarget:$offset) // (ins IntRegs:$src1, u5Imm:$src2, brtarget:$offset) unsigned Flags1 = getUndefRegState(Cond[1].isUndef()); LLVM_DEBUG(dbgs() << "\nInserting NVJump for " << printMBBReference(MBB);); if (Cond[2].isReg()) { unsigned Flags2 = getUndefRegState(Cond[2].isUndef()); BuildMI(&MBB, DL, get(BccOpc)).addReg(Cond[1].getReg(), Flags1). addReg(Cond[2].getReg(), Flags2).addMBB(TBB); } else if(Cond[2].isImm()) { BuildMI(&MBB, DL, get(BccOpc)).addReg(Cond[1].getReg(), Flags1). addImm(Cond[2].getImm()).addMBB(TBB); } else llvm_unreachable("Invalid condition for branching"); } else { assert((Cond.size() == 2) && "Malformed cond vector"); const MachineOperand &RO = Cond[1]; unsigned Flags = getUndefRegState(RO.isUndef()); BuildMI(&MBB, DL, get(BccOpc)).addReg(RO.getReg(), Flags).addMBB(TBB); } return 1; } assert((!Cond.empty()) && "Cond. cannot be empty when multiple branchings are required"); assert((!isNewValueJump(Cond[0].getImm())) && "NV-jump cannot be inserted with another branch"); // Special case for hardware loops. The condition is a basic block. if (isEndLoopN(Cond[0].getImm())) { int EndLoopOp = Cond[0].getImm(); assert(Cond[1].isMBB()); // Since we're adding an ENDLOOP, there better be a LOOP instruction. // Check for it, and change the BB target if needed. SmallPtrSet VisitedBBs; MachineInstr *Loop = findLoopInstr(TBB, EndLoopOp, Cond[1].getMBB(), VisitedBBs); assert(Loop != nullptr && "Inserting an ENDLOOP without a LOOP"); Loop->getOperand(0).setMBB(TBB); // Add the ENDLOOP after the finding the LOOP0. BuildMI(&MBB, DL, get(EndLoopOp)).addMBB(TBB); } else { const MachineOperand &RO = Cond[1]; unsigned Flags = getUndefRegState(RO.isUndef()); BuildMI(&MBB, DL, get(BccOpc)).addReg(RO.getReg(), Flags).addMBB(TBB); } BuildMI(&MBB, DL, get(BOpc)).addMBB(FBB); return 2; } namespace { class HexagonPipelinerLoopInfo : public TargetInstrInfo::PipelinerLoopInfo { MachineInstr *Loop, *EndLoop; MachineFunction *MF; const HexagonInstrInfo *TII; int64_t TripCount; Register LoopCount; DebugLoc DL; public: HexagonPipelinerLoopInfo(MachineInstr *Loop, MachineInstr *EndLoop) : Loop(Loop), EndLoop(EndLoop), MF(Loop->getParent()->getParent()), TII(MF->getSubtarget().getInstrInfo()), DL(Loop->getDebugLoc()) { // Inspect the Loop instruction up-front, as it may be deleted when we call // createTripCountGreaterCondition. TripCount = Loop->getOpcode() == Hexagon::J2_loop0r ? -1 : Loop->getOperand(1).getImm(); if (TripCount == -1) LoopCount = Loop->getOperand(1).getReg(); } bool shouldIgnoreForPipelining(const MachineInstr *MI) const override { // Only ignore the terminator. return MI == EndLoop; } Optional createTripCountGreaterCondition(int TC, MachineBasicBlock &MBB, SmallVectorImpl &Cond) override { if (TripCount == -1) { // Check if we're done with the loop. unsigned Done = TII->createVR(MF, MVT::i1); MachineInstr *NewCmp = BuildMI(&MBB, DL, TII->get(Hexagon::C2_cmpgtui), Done) .addReg(LoopCount) .addImm(TC); Cond.push_back(MachineOperand::CreateImm(Hexagon::J2_jumpf)); Cond.push_back(NewCmp->getOperand(0)); return {}; } return TripCount > TC; } void setPreheader(MachineBasicBlock *NewPreheader) override { NewPreheader->splice(NewPreheader->getFirstTerminator(), Loop->getParent(), Loop); } void adjustTripCount(int TripCountAdjust) override { // If the loop trip count is a compile-time value, then just change the // value. if (Loop->getOpcode() == Hexagon::J2_loop0i || Loop->getOpcode() == Hexagon::J2_loop1i) { int64_t TripCount = Loop->getOperand(1).getImm() + TripCountAdjust; assert(TripCount > 0 && "Can't create an empty or negative loop!"); Loop->getOperand(1).setImm(TripCount); return; } // The loop trip count is a run-time value. We generate code to subtract // one from the trip count, and update the loop instruction. Register LoopCount = Loop->getOperand(1).getReg(); Register NewLoopCount = TII->createVR(MF, MVT::i32); BuildMI(*Loop->getParent(), Loop, Loop->getDebugLoc(), TII->get(Hexagon::A2_addi), NewLoopCount) .addReg(LoopCount) .addImm(TripCountAdjust); Loop->getOperand(1).setReg(NewLoopCount); } void disposed() override { Loop->eraseFromParent(); } }; } // namespace std::unique_ptr HexagonInstrInfo::analyzeLoopForPipelining(MachineBasicBlock *LoopBB) const { // We really "analyze" only hardware loops right now. MachineBasicBlock::iterator I = LoopBB->getFirstTerminator(); if (I != LoopBB->end() && isEndLoopN(I->getOpcode())) { SmallPtrSet VisitedBBs; MachineInstr *LoopInst = findLoopInstr( LoopBB, I->getOpcode(), I->getOperand(0).getMBB(), VisitedBBs); if (LoopInst) return std::make_unique(LoopInst, &*I); } return nullptr; } bool HexagonInstrInfo::isProfitableToIfCvt(MachineBasicBlock &MBB, unsigned NumCycles, unsigned ExtraPredCycles, BranchProbability Probability) const { return nonDbgBBSize(&MBB) <= 3; } bool HexagonInstrInfo::isProfitableToIfCvt(MachineBasicBlock &TMBB, unsigned NumTCycles, unsigned ExtraTCycles, MachineBasicBlock &FMBB, unsigned NumFCycles, unsigned ExtraFCycles, BranchProbability Probability) const { return nonDbgBBSize(&TMBB) <= 3 && nonDbgBBSize(&FMBB) <= 3; } bool HexagonInstrInfo::isProfitableToDupForIfCvt(MachineBasicBlock &MBB, unsigned NumInstrs, BranchProbability Probability) const { return NumInstrs <= 4; } static void getLiveInRegsAt(LivePhysRegs &Regs, const MachineInstr &MI) { SmallVector,2> Clobbers; const MachineBasicBlock &B = *MI.getParent(); Regs.addLiveIns(B); auto E = MachineBasicBlock::const_iterator(MI.getIterator()); for (auto I = B.begin(); I != E; ++I) { Clobbers.clear(); Regs.stepForward(*I, Clobbers); } } static void getLiveOutRegsAt(LivePhysRegs &Regs, const MachineInstr &MI) { const MachineBasicBlock &B = *MI.getParent(); Regs.addLiveOuts(B); auto E = ++MachineBasicBlock::const_iterator(MI.getIterator()).getReverse(); for (auto I = B.rbegin(); I != E; ++I) Regs.stepBackward(*I); } void HexagonInstrInfo::copyPhysReg(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, const DebugLoc &DL, MCRegister DestReg, MCRegister SrcReg, bool KillSrc) const { const HexagonRegisterInfo &HRI = *Subtarget.getRegisterInfo(); unsigned KillFlag = getKillRegState(KillSrc); if (Hexagon::IntRegsRegClass.contains(SrcReg, DestReg)) { BuildMI(MBB, I, DL, get(Hexagon::A2_tfr), DestReg) .addReg(SrcReg, KillFlag); return; } if (Hexagon::DoubleRegsRegClass.contains(SrcReg, DestReg)) { BuildMI(MBB, I, DL, get(Hexagon::A2_tfrp), DestReg) .addReg(SrcReg, KillFlag); return; } if (Hexagon::PredRegsRegClass.contains(SrcReg, DestReg)) { // Map Pd = Ps to Pd = or(Ps, Ps). BuildMI(MBB, I, DL, get(Hexagon::C2_or), DestReg) .addReg(SrcReg).addReg(SrcReg, KillFlag); return; } if (Hexagon::CtrRegsRegClass.contains(DestReg) && Hexagon::IntRegsRegClass.contains(SrcReg)) { BuildMI(MBB, I, DL, get(Hexagon::A2_tfrrcr), DestReg) .addReg(SrcReg, KillFlag); return; } if (Hexagon::IntRegsRegClass.contains(DestReg) && Hexagon::CtrRegsRegClass.contains(SrcReg)) { BuildMI(MBB, I, DL, get(Hexagon::A2_tfrcrr), DestReg) .addReg(SrcReg, KillFlag); return; } if (Hexagon::ModRegsRegClass.contains(DestReg) && Hexagon::IntRegsRegClass.contains(SrcReg)) { BuildMI(MBB, I, DL, get(Hexagon::A2_tfrrcr), DestReg) .addReg(SrcReg, KillFlag); return; } if (Hexagon::PredRegsRegClass.contains(SrcReg) && Hexagon::IntRegsRegClass.contains(DestReg)) { BuildMI(MBB, I, DL, get(Hexagon::C2_tfrpr), DestReg) .addReg(SrcReg, KillFlag); return; } if (Hexagon::IntRegsRegClass.contains(SrcReg) && Hexagon::PredRegsRegClass.contains(DestReg)) { BuildMI(MBB, I, DL, get(Hexagon::C2_tfrrp), DestReg) .addReg(SrcReg, KillFlag); return; } if (Hexagon::PredRegsRegClass.contains(SrcReg) && Hexagon::IntRegsRegClass.contains(DestReg)) { BuildMI(MBB, I, DL, get(Hexagon::C2_tfrpr), DestReg) .addReg(SrcReg, KillFlag); return; } if (Hexagon::HvxVRRegClass.contains(SrcReg, DestReg)) { BuildMI(MBB, I, DL, get(Hexagon::V6_vassign), DestReg). addReg(SrcReg, KillFlag); return; } if (Hexagon::HvxWRRegClass.contains(SrcReg, DestReg)) { LivePhysRegs LiveAtMI(HRI); getLiveInRegsAt(LiveAtMI, *I); Register SrcLo = HRI.getSubReg(SrcReg, Hexagon::vsub_lo); Register SrcHi = HRI.getSubReg(SrcReg, Hexagon::vsub_hi); unsigned UndefLo = getUndefRegState(!LiveAtMI.contains(SrcLo)); unsigned UndefHi = getUndefRegState(!LiveAtMI.contains(SrcHi)); BuildMI(MBB, I, DL, get(Hexagon::V6_vcombine), DestReg) .addReg(SrcHi, KillFlag | UndefHi) .addReg(SrcLo, KillFlag | UndefLo); return; } if (Hexagon::HvxQRRegClass.contains(SrcReg, DestReg)) { BuildMI(MBB, I, DL, get(Hexagon::V6_pred_and), DestReg) .addReg(SrcReg) .addReg(SrcReg, KillFlag); return; } if (Hexagon::HvxQRRegClass.contains(SrcReg) && Hexagon::HvxVRRegClass.contains(DestReg)) { llvm_unreachable("Unimplemented pred to vec"); return; } if (Hexagon::HvxQRRegClass.contains(DestReg) && Hexagon::HvxVRRegClass.contains(SrcReg)) { llvm_unreachable("Unimplemented vec to pred"); return; } #ifndef NDEBUG // Show the invalid registers to ease debugging. dbgs() << "Invalid registers for copy in " << printMBBReference(MBB) << ": " << printReg(DestReg, &HRI) << " = " << printReg(SrcReg, &HRI) << '\n'; #endif llvm_unreachable("Unimplemented"); } void HexagonInstrInfo::storeRegToStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, Register SrcReg, bool isKill, int FI, const TargetRegisterClass *RC, const TargetRegisterInfo *TRI) const { DebugLoc DL = MBB.findDebugLoc(I); MachineFunction &MF = *MBB.getParent(); MachineFrameInfo &MFI = MF.getFrameInfo(); unsigned KillFlag = getKillRegState(isKill); MachineMemOperand *MMO = MF.getMachineMemOperand( MachinePointerInfo::getFixedStack(MF, FI), MachineMemOperand::MOStore, MFI.getObjectSize(FI), MFI.getObjectAlign(FI)); if (Hexagon::IntRegsRegClass.hasSubClassEq(RC)) { BuildMI(MBB, I, DL, get(Hexagon::S2_storeri_io)) .addFrameIndex(FI).addImm(0) .addReg(SrcReg, KillFlag).addMemOperand(MMO); } else if (Hexagon::DoubleRegsRegClass.hasSubClassEq(RC)) { BuildMI(MBB, I, DL, get(Hexagon::S2_storerd_io)) .addFrameIndex(FI).addImm(0) .addReg(SrcReg, KillFlag).addMemOperand(MMO); } else if (Hexagon::PredRegsRegClass.hasSubClassEq(RC)) { BuildMI(MBB, I, DL, get(Hexagon::STriw_pred)) .addFrameIndex(FI).addImm(0) .addReg(SrcReg, KillFlag).addMemOperand(MMO); } else if (Hexagon::ModRegsRegClass.hasSubClassEq(RC)) { BuildMI(MBB, I, DL, get(Hexagon::STriw_ctr)) .addFrameIndex(FI).addImm(0) .addReg(SrcReg, KillFlag).addMemOperand(MMO); } else if (Hexagon::HvxQRRegClass.hasSubClassEq(RC)) { BuildMI(MBB, I, DL, get(Hexagon::PS_vstorerq_ai)) .addFrameIndex(FI).addImm(0) .addReg(SrcReg, KillFlag).addMemOperand(MMO); } else if (Hexagon::HvxVRRegClass.hasSubClassEq(RC)) { BuildMI(MBB, I, DL, get(Hexagon::PS_vstorerv_ai)) .addFrameIndex(FI).addImm(0) .addReg(SrcReg, KillFlag).addMemOperand(MMO); } else if (Hexagon::HvxWRRegClass.hasSubClassEq(RC)) { BuildMI(MBB, I, DL, get(Hexagon::PS_vstorerw_ai)) .addFrameIndex(FI).addImm(0) .addReg(SrcReg, KillFlag).addMemOperand(MMO); } else { llvm_unreachable("Unimplemented"); } } void HexagonInstrInfo::loadRegFromStackSlot( MachineBasicBlock &MBB, MachineBasicBlock::iterator I, Register DestReg, int FI, const TargetRegisterClass *RC, const TargetRegisterInfo *TRI) const { DebugLoc DL = MBB.findDebugLoc(I); MachineFunction &MF = *MBB.getParent(); MachineFrameInfo &MFI = MF.getFrameInfo(); MachineMemOperand *MMO = MF.getMachineMemOperand( MachinePointerInfo::getFixedStack(MF, FI), MachineMemOperand::MOLoad, MFI.getObjectSize(FI), MFI.getObjectAlign(FI)); if (Hexagon::IntRegsRegClass.hasSubClassEq(RC)) { BuildMI(MBB, I, DL, get(Hexagon::L2_loadri_io), DestReg) .addFrameIndex(FI).addImm(0).addMemOperand(MMO); } else if (Hexagon::DoubleRegsRegClass.hasSubClassEq(RC)) { BuildMI(MBB, I, DL, get(Hexagon::L2_loadrd_io), DestReg) .addFrameIndex(FI).addImm(0).addMemOperand(MMO); } else if (Hexagon::PredRegsRegClass.hasSubClassEq(RC)) { BuildMI(MBB, I, DL, get(Hexagon::LDriw_pred), DestReg) .addFrameIndex(FI).addImm(0).addMemOperand(MMO); } else if (Hexagon::ModRegsRegClass.hasSubClassEq(RC)) { BuildMI(MBB, I, DL, get(Hexagon::LDriw_ctr), DestReg) .addFrameIndex(FI).addImm(0).addMemOperand(MMO); } else if (Hexagon::HvxQRRegClass.hasSubClassEq(RC)) { BuildMI(MBB, I, DL, get(Hexagon::PS_vloadrq_ai), DestReg) .addFrameIndex(FI).addImm(0).addMemOperand(MMO); } else if (Hexagon::HvxVRRegClass.hasSubClassEq(RC)) { BuildMI(MBB, I, DL, get(Hexagon::PS_vloadrv_ai), DestReg) .addFrameIndex(FI).addImm(0).addMemOperand(MMO); } else if (Hexagon::HvxWRRegClass.hasSubClassEq(RC)) { BuildMI(MBB, I, DL, get(Hexagon::PS_vloadrw_ai), DestReg) .addFrameIndex(FI).addImm(0).addMemOperand(MMO); } else { llvm_unreachable("Can't store this register to stack slot"); } } /// expandPostRAPseudo - This function is called for all pseudo instructions /// that remain after register allocation. Many pseudo instructions are /// created to help register allocation. This is the place to convert them /// into real instructions. The target can edit MI in place, or it can insert /// new instructions and erase MI. The function should return true if /// anything was changed. bool HexagonInstrInfo::expandPostRAPseudo(MachineInstr &MI) const { MachineBasicBlock &MBB = *MI.getParent(); MachineFunction &MF = *MBB.getParent(); MachineRegisterInfo &MRI = MF.getRegInfo(); const HexagonRegisterInfo &HRI = *Subtarget.getRegisterInfo(); LivePhysRegs LiveIn(HRI), LiveOut(HRI); DebugLoc DL = MI.getDebugLoc(); unsigned Opc = MI.getOpcode(); auto RealCirc = [&](unsigned Opc, bool HasImm, unsigned MxOp) { Register Mx = MI.getOperand(MxOp).getReg(); unsigned CSx = (Mx == Hexagon::M0 ? Hexagon::CS0 : Hexagon::CS1); BuildMI(MBB, MI, DL, get(Hexagon::A2_tfrrcr), CSx) .add(MI.getOperand((HasImm ? 5 : 4))); auto MIB = BuildMI(MBB, MI, DL, get(Opc)).add(MI.getOperand(0)) .add(MI.getOperand(1)).add(MI.getOperand(2)).add(MI.getOperand(3)); if (HasImm) MIB.add(MI.getOperand(4)); MIB.addReg(CSx, RegState::Implicit); MBB.erase(MI); return true; }; auto UseAligned = [&](const MachineInstr &MI, Align NeedAlign) { if (MI.memoperands().empty()) return false; return all_of(MI.memoperands(), [NeedAlign](const MachineMemOperand *MMO) { return MMO->getAlign() >= NeedAlign; }); }; switch (Opc) { case TargetOpcode::COPY: { MachineOperand &MD = MI.getOperand(0); MachineOperand &MS = MI.getOperand(1); MachineBasicBlock::iterator MBBI = MI.getIterator(); if (MD.getReg() != MS.getReg() && !MS.isUndef()) { copyPhysReg(MBB, MI, DL, MD.getReg(), MS.getReg(), MS.isKill()); std::prev(MBBI)->copyImplicitOps(*MBB.getParent(), MI); } MBB.erase(MBBI); return true; } case Hexagon::PS_aligna: BuildMI(MBB, MI, DL, get(Hexagon::A2_andir), MI.getOperand(0).getReg()) .addReg(HRI.getFrameRegister()) .addImm(-MI.getOperand(1).getImm()); MBB.erase(MI); return true; case Hexagon::V6_vassignp: { Register SrcReg = MI.getOperand(1).getReg(); Register DstReg = MI.getOperand(0).getReg(); Register SrcLo = HRI.getSubReg(SrcReg, Hexagon::vsub_lo); Register SrcHi = HRI.getSubReg(SrcReg, Hexagon::vsub_hi); getLiveInRegsAt(LiveIn, MI); unsigned UndefLo = getUndefRegState(!LiveIn.contains(SrcLo)); unsigned UndefHi = getUndefRegState(!LiveIn.contains(SrcHi)); unsigned Kill = getKillRegState(MI.getOperand(1).isKill()); BuildMI(MBB, MI, DL, get(Hexagon::V6_vcombine), DstReg) .addReg(SrcHi, UndefHi) .addReg(SrcLo, Kill | UndefLo); MBB.erase(MI); return true; } case Hexagon::V6_lo: { Register SrcReg = MI.getOperand(1).getReg(); Register DstReg = MI.getOperand(0).getReg(); Register SrcSubLo = HRI.getSubReg(SrcReg, Hexagon::vsub_lo); copyPhysReg(MBB, MI, DL, DstReg, SrcSubLo, MI.getOperand(1).isKill()); MBB.erase(MI); MRI.clearKillFlags(SrcSubLo); return true; } case Hexagon::V6_hi: { Register SrcReg = MI.getOperand(1).getReg(); Register DstReg = MI.getOperand(0).getReg(); Register SrcSubHi = HRI.getSubReg(SrcReg, Hexagon::vsub_hi); copyPhysReg(MBB, MI, DL, DstReg, SrcSubHi, MI.getOperand(1).isKill()); MBB.erase(MI); MRI.clearKillFlags(SrcSubHi); return true; } case Hexagon::PS_vloadrv_ai: { Register DstReg = MI.getOperand(0).getReg(); const MachineOperand &BaseOp = MI.getOperand(1); assert(BaseOp.getSubReg() == 0); int Offset = MI.getOperand(2).getImm(); Align NeedAlign = HRI.getSpillAlign(Hexagon::HvxVRRegClass); unsigned NewOpc = UseAligned(MI, NeedAlign) ? Hexagon::V6_vL32b_ai : Hexagon::V6_vL32Ub_ai; BuildMI(MBB, MI, DL, get(NewOpc), DstReg) .addReg(BaseOp.getReg(), getRegState(BaseOp)) .addImm(Offset) .cloneMemRefs(MI); MBB.erase(MI); return true; } case Hexagon::PS_vloadrw_ai: { Register DstReg = MI.getOperand(0).getReg(); const MachineOperand &BaseOp = MI.getOperand(1); assert(BaseOp.getSubReg() == 0); int Offset = MI.getOperand(2).getImm(); unsigned VecOffset = HRI.getSpillSize(Hexagon::HvxVRRegClass); Align NeedAlign = HRI.getSpillAlign(Hexagon::HvxVRRegClass); unsigned NewOpc = UseAligned(MI, NeedAlign) ? Hexagon::V6_vL32b_ai : Hexagon::V6_vL32Ub_ai; BuildMI(MBB, MI, DL, get(NewOpc), HRI.getSubReg(DstReg, Hexagon::vsub_lo)) .addReg(BaseOp.getReg(), getRegState(BaseOp) & ~RegState::Kill) .addImm(Offset) .cloneMemRefs(MI); BuildMI(MBB, MI, DL, get(NewOpc), HRI.getSubReg(DstReg, Hexagon::vsub_hi)) .addReg(BaseOp.getReg(), getRegState(BaseOp)) .addImm(Offset + VecOffset) .cloneMemRefs(MI); MBB.erase(MI); return true; } case Hexagon::PS_vstorerv_ai: { const MachineOperand &SrcOp = MI.getOperand(2); assert(SrcOp.getSubReg() == 0); const MachineOperand &BaseOp = MI.getOperand(0); assert(BaseOp.getSubReg() == 0); int Offset = MI.getOperand(1).getImm(); Align NeedAlign = HRI.getSpillAlign(Hexagon::HvxVRRegClass); unsigned NewOpc = UseAligned(MI, NeedAlign) ? Hexagon::V6_vS32b_ai : Hexagon::V6_vS32Ub_ai; BuildMI(MBB, MI, DL, get(NewOpc)) .addReg(BaseOp.getReg(), getRegState(BaseOp)) .addImm(Offset) .addReg(SrcOp.getReg(), getRegState(SrcOp)) .cloneMemRefs(MI); MBB.erase(MI); return true; } case Hexagon::PS_vstorerw_ai: { Register SrcReg = MI.getOperand(2).getReg(); const MachineOperand &BaseOp = MI.getOperand(0); assert(BaseOp.getSubReg() == 0); int Offset = MI.getOperand(1).getImm(); unsigned VecOffset = HRI.getSpillSize(Hexagon::HvxVRRegClass); Align NeedAlign = HRI.getSpillAlign(Hexagon::HvxVRRegClass); unsigned NewOpc = UseAligned(MI, NeedAlign) ? Hexagon::V6_vS32b_ai : Hexagon::V6_vS32Ub_ai; BuildMI(MBB, MI, DL, get(NewOpc)) .addReg(BaseOp.getReg(), getRegState(BaseOp) & ~RegState::Kill) .addImm(Offset) .addReg(HRI.getSubReg(SrcReg, Hexagon::vsub_lo)) .cloneMemRefs(MI); BuildMI(MBB, MI, DL, get(NewOpc)) .addReg(BaseOp.getReg(), getRegState(BaseOp)) .addImm(Offset + VecOffset) .addReg(HRI.getSubReg(SrcReg, Hexagon::vsub_hi)) .cloneMemRefs(MI); MBB.erase(MI); return true; } case Hexagon::PS_true: { Register Reg = MI.getOperand(0).getReg(); BuildMI(MBB, MI, DL, get(Hexagon::C2_orn), Reg) .addReg(Reg, RegState::Undef) .addReg(Reg, RegState::Undef); MBB.erase(MI); return true; } case Hexagon::PS_false: { Register Reg = MI.getOperand(0).getReg(); BuildMI(MBB, MI, DL, get(Hexagon::C2_andn), Reg) .addReg(Reg, RegState::Undef) .addReg(Reg, RegState::Undef); MBB.erase(MI); return true; } case Hexagon::PS_qtrue: { BuildMI(MBB, MI, DL, get(Hexagon::V6_veqw), MI.getOperand(0).getReg()) .addReg(Hexagon::V0, RegState::Undef) .addReg(Hexagon::V0, RegState::Undef); MBB.erase(MI); return true; } case Hexagon::PS_qfalse: { BuildMI(MBB, MI, DL, get(Hexagon::V6_vgtw), MI.getOperand(0).getReg()) .addReg(Hexagon::V0, RegState::Undef) .addReg(Hexagon::V0, RegState::Undef); MBB.erase(MI); return true; } case Hexagon::PS_vdd0: { Register Vd = MI.getOperand(0).getReg(); BuildMI(MBB, MI, DL, get(Hexagon::V6_vsubw_dv), Vd) .addReg(Vd, RegState::Undef) .addReg(Vd, RegState::Undef); MBB.erase(MI); return true; } case Hexagon::PS_vmulw: { // Expand a 64-bit vector multiply into 2 32-bit scalar multiplies. Register DstReg = MI.getOperand(0).getReg(); Register Src1Reg = MI.getOperand(1).getReg(); Register Src2Reg = MI.getOperand(2).getReg(); Register Src1SubHi = HRI.getSubReg(Src1Reg, Hexagon::isub_hi); Register Src1SubLo = HRI.getSubReg(Src1Reg, Hexagon::isub_lo); Register Src2SubHi = HRI.getSubReg(Src2Reg, Hexagon::isub_hi); Register Src2SubLo = HRI.getSubReg(Src2Reg, Hexagon::isub_lo); BuildMI(MBB, MI, MI.getDebugLoc(), get(Hexagon::M2_mpyi), HRI.getSubReg(DstReg, Hexagon::isub_hi)) .addReg(Src1SubHi) .addReg(Src2SubHi); BuildMI(MBB, MI, MI.getDebugLoc(), get(Hexagon::M2_mpyi), HRI.getSubReg(DstReg, Hexagon::isub_lo)) .addReg(Src1SubLo) .addReg(Src2SubLo); MBB.erase(MI); MRI.clearKillFlags(Src1SubHi); MRI.clearKillFlags(Src1SubLo); MRI.clearKillFlags(Src2SubHi); MRI.clearKillFlags(Src2SubLo); return true; } case Hexagon::PS_vmulw_acc: { // Expand 64-bit vector multiply with addition into 2 scalar multiplies. Register DstReg = MI.getOperand(0).getReg(); Register Src1Reg = MI.getOperand(1).getReg(); Register Src2Reg = MI.getOperand(2).getReg(); Register Src3Reg = MI.getOperand(3).getReg(); Register Src1SubHi = HRI.getSubReg(Src1Reg, Hexagon::isub_hi); Register Src1SubLo = HRI.getSubReg(Src1Reg, Hexagon::isub_lo); Register Src2SubHi = HRI.getSubReg(Src2Reg, Hexagon::isub_hi); Register Src2SubLo = HRI.getSubReg(Src2Reg, Hexagon::isub_lo); Register Src3SubHi = HRI.getSubReg(Src3Reg, Hexagon::isub_hi); Register Src3SubLo = HRI.getSubReg(Src3Reg, Hexagon::isub_lo); BuildMI(MBB, MI, MI.getDebugLoc(), get(Hexagon::M2_maci), HRI.getSubReg(DstReg, Hexagon::isub_hi)) .addReg(Src1SubHi) .addReg(Src2SubHi) .addReg(Src3SubHi); BuildMI(MBB, MI, MI.getDebugLoc(), get(Hexagon::M2_maci), HRI.getSubReg(DstReg, Hexagon::isub_lo)) .addReg(Src1SubLo) .addReg(Src2SubLo) .addReg(Src3SubLo); MBB.erase(MI); MRI.clearKillFlags(Src1SubHi); MRI.clearKillFlags(Src1SubLo); MRI.clearKillFlags(Src2SubHi); MRI.clearKillFlags(Src2SubLo); MRI.clearKillFlags(Src3SubHi); MRI.clearKillFlags(Src3SubLo); return true; } case Hexagon::PS_pselect: { const MachineOperand &Op0 = MI.getOperand(0); const MachineOperand &Op1 = MI.getOperand(1); const MachineOperand &Op2 = MI.getOperand(2); const MachineOperand &Op3 = MI.getOperand(3); Register Rd = Op0.getReg(); Register Pu = Op1.getReg(); Register Rs = Op2.getReg(); Register Rt = Op3.getReg(); DebugLoc DL = MI.getDebugLoc(); unsigned K1 = getKillRegState(Op1.isKill()); unsigned K2 = getKillRegState(Op2.isKill()); unsigned K3 = getKillRegState(Op3.isKill()); if (Rd != Rs) BuildMI(MBB, MI, DL, get(Hexagon::A2_tfrpt), Rd) .addReg(Pu, (Rd == Rt) ? K1 : 0) .addReg(Rs, K2); if (Rd != Rt) BuildMI(MBB, MI, DL, get(Hexagon::A2_tfrpf), Rd) .addReg(Pu, K1) .addReg(Rt, K3); MBB.erase(MI); return true; } case Hexagon::PS_vselect: { const MachineOperand &Op0 = MI.getOperand(0); const MachineOperand &Op1 = MI.getOperand(1); const MachineOperand &Op2 = MI.getOperand(2); const MachineOperand &Op3 = MI.getOperand(3); getLiveOutRegsAt(LiveOut, MI); bool IsDestLive = !LiveOut.available(MRI, Op0.getReg()); Register PReg = Op1.getReg(); assert(Op1.getSubReg() == 0); unsigned PState = getRegState(Op1); if (Op0.getReg() != Op2.getReg()) { unsigned S = Op0.getReg() != Op3.getReg() ? PState & ~RegState::Kill : PState; auto T = BuildMI(MBB, MI, DL, get(Hexagon::V6_vcmov)) .add(Op0) .addReg(PReg, S) .add(Op2); if (IsDestLive) T.addReg(Op0.getReg(), RegState::Implicit); IsDestLive = true; } if (Op0.getReg() != Op3.getReg()) { auto T = BuildMI(MBB, MI, DL, get(Hexagon::V6_vncmov)) .add(Op0) .addReg(PReg, PState) .add(Op3); if (IsDestLive) T.addReg(Op0.getReg(), RegState::Implicit); } MBB.erase(MI); return true; } case Hexagon::PS_wselect: { MachineOperand &Op0 = MI.getOperand(0); MachineOperand &Op1 = MI.getOperand(1); MachineOperand &Op2 = MI.getOperand(2); MachineOperand &Op3 = MI.getOperand(3); getLiveOutRegsAt(LiveOut, MI); bool IsDestLive = !LiveOut.available(MRI, Op0.getReg()); Register PReg = Op1.getReg(); assert(Op1.getSubReg() == 0); unsigned PState = getRegState(Op1); if (Op0.getReg() != Op2.getReg()) { unsigned S = Op0.getReg() != Op3.getReg() ? PState & ~RegState::Kill : PState; Register SrcLo = HRI.getSubReg(Op2.getReg(), Hexagon::vsub_lo); Register SrcHi = HRI.getSubReg(Op2.getReg(), Hexagon::vsub_hi); auto T = BuildMI(MBB, MI, DL, get(Hexagon::V6_vccombine)) .add(Op0) .addReg(PReg, S) .addReg(SrcHi) .addReg(SrcLo); if (IsDestLive) T.addReg(Op0.getReg(), RegState::Implicit); IsDestLive = true; } if (Op0.getReg() != Op3.getReg()) { Register SrcLo = HRI.getSubReg(Op3.getReg(), Hexagon::vsub_lo); Register SrcHi = HRI.getSubReg(Op3.getReg(), Hexagon::vsub_hi); auto T = BuildMI(MBB, MI, DL, get(Hexagon::V6_vnccombine)) .add(Op0) .addReg(PReg, PState) .addReg(SrcHi) .addReg(SrcLo); if (IsDestLive) T.addReg(Op0.getReg(), RegState::Implicit); } MBB.erase(MI); return true; } case Hexagon::PS_crash: { // Generate a misaligned load that is guaranteed to cause a crash. class CrashPseudoSourceValue : public PseudoSourceValue { public: CrashPseudoSourceValue(const TargetInstrInfo &TII) : PseudoSourceValue(TargetCustom, TII) {} bool isConstant(const MachineFrameInfo *) const override { return false; } bool isAliased(const MachineFrameInfo *) const override { return false; } bool mayAlias(const MachineFrameInfo *) const override { return false; } void printCustom(raw_ostream &OS) const override { OS << "MisalignedCrash"; } }; static const CrashPseudoSourceValue CrashPSV(*this); MachineMemOperand *MMO = MF.getMachineMemOperand( MachinePointerInfo(&CrashPSV), MachineMemOperand::MOLoad | MachineMemOperand::MOVolatile, 8, Align(1)); BuildMI(MBB, MI, DL, get(Hexagon::PS_loadrdabs), Hexagon::D13) .addImm(0xBADC0FEE) // Misaligned load. .addMemOperand(MMO); MBB.erase(MI); return true; } case Hexagon::PS_tailcall_i: MI.setDesc(get(Hexagon::J2_jump)); return true; case Hexagon::PS_tailcall_r: case Hexagon::PS_jmpret: MI.setDesc(get(Hexagon::J2_jumpr)); return true; case Hexagon::PS_jmprett: MI.setDesc(get(Hexagon::J2_jumprt)); return true; case Hexagon::PS_jmpretf: MI.setDesc(get(Hexagon::J2_jumprf)); return true; case Hexagon::PS_jmprettnewpt: MI.setDesc(get(Hexagon::J2_jumprtnewpt)); return true; case Hexagon::PS_jmpretfnewpt: MI.setDesc(get(Hexagon::J2_jumprfnewpt)); return true; case Hexagon::PS_jmprettnew: MI.setDesc(get(Hexagon::J2_jumprtnew)); return true; case Hexagon::PS_jmpretfnew: MI.setDesc(get(Hexagon::J2_jumprfnew)); return true; case Hexagon::PS_loadrub_pci: return RealCirc(Hexagon::L2_loadrub_pci, /*HasImm*/true, /*MxOp*/4); case Hexagon::PS_loadrb_pci: return RealCirc(Hexagon::L2_loadrb_pci, /*HasImm*/true, /*MxOp*/4); case Hexagon::PS_loadruh_pci: return RealCirc(Hexagon::L2_loadruh_pci, /*HasImm*/true, /*MxOp*/4); case Hexagon::PS_loadrh_pci: return RealCirc(Hexagon::L2_loadrh_pci, /*HasImm*/true, /*MxOp*/4); case Hexagon::PS_loadri_pci: return RealCirc(Hexagon::L2_loadri_pci, /*HasImm*/true, /*MxOp*/4); case Hexagon::PS_loadrd_pci: return RealCirc(Hexagon::L2_loadrd_pci, /*HasImm*/true, /*MxOp*/4); case Hexagon::PS_loadrub_pcr: return RealCirc(Hexagon::L2_loadrub_pcr, /*HasImm*/false, /*MxOp*/3); case Hexagon::PS_loadrb_pcr: return RealCirc(Hexagon::L2_loadrb_pcr, /*HasImm*/false, /*MxOp*/3); case Hexagon::PS_loadruh_pcr: return RealCirc(Hexagon::L2_loadruh_pcr, /*HasImm*/false, /*MxOp*/3); case Hexagon::PS_loadrh_pcr: return RealCirc(Hexagon::L2_loadrh_pcr, /*HasImm*/false, /*MxOp*/3); case Hexagon::PS_loadri_pcr: return RealCirc(Hexagon::L2_loadri_pcr, /*HasImm*/false, /*MxOp*/3); case Hexagon::PS_loadrd_pcr: return RealCirc(Hexagon::L2_loadrd_pcr, /*HasImm*/false, /*MxOp*/3); case Hexagon::PS_storerb_pci: return RealCirc(Hexagon::S2_storerb_pci, /*HasImm*/true, /*MxOp*/3); case Hexagon::PS_storerh_pci: return RealCirc(Hexagon::S2_storerh_pci, /*HasImm*/true, /*MxOp*/3); case Hexagon::PS_storerf_pci: return RealCirc(Hexagon::S2_storerf_pci, /*HasImm*/true, /*MxOp*/3); case Hexagon::PS_storeri_pci: return RealCirc(Hexagon::S2_storeri_pci, /*HasImm*/true, /*MxOp*/3); case Hexagon::PS_storerd_pci: return RealCirc(Hexagon::S2_storerd_pci, /*HasImm*/true, /*MxOp*/3); case Hexagon::PS_storerb_pcr: return RealCirc(Hexagon::S2_storerb_pcr, /*HasImm*/false, /*MxOp*/2); case Hexagon::PS_storerh_pcr: return RealCirc(Hexagon::S2_storerh_pcr, /*HasImm*/false, /*MxOp*/2); case Hexagon::PS_storerf_pcr: return RealCirc(Hexagon::S2_storerf_pcr, /*HasImm*/false, /*MxOp*/2); case Hexagon::PS_storeri_pcr: return RealCirc(Hexagon::S2_storeri_pcr, /*HasImm*/false, /*MxOp*/2); case Hexagon::PS_storerd_pcr: return RealCirc(Hexagon::S2_storerd_pcr, /*HasImm*/false, /*MxOp*/2); } return false; } MachineBasicBlock::instr_iterator HexagonInstrInfo::expandVGatherPseudo(MachineInstr &MI) const { MachineBasicBlock &MBB = *MI.getParent(); const DebugLoc &DL = MI.getDebugLoc(); unsigned Opc = MI.getOpcode(); MachineBasicBlock::iterator First; switch (Opc) { case Hexagon::V6_vgathermh_pseudo: First = BuildMI(MBB, MI, DL, get(Hexagon::V6_vgathermh)) .add(MI.getOperand(1)) .add(MI.getOperand(2)) .add(MI.getOperand(3)); BuildMI(MBB, MI, DL, get(Hexagon::V6_vS32b_new_ai)) .add(MI.getOperand(0)) .addImm(0) .addReg(Hexagon::VTMP); MBB.erase(MI); return First.getInstrIterator(); case Hexagon::V6_vgathermw_pseudo: First = BuildMI(MBB, MI, DL, get(Hexagon::V6_vgathermw)) .add(MI.getOperand(1)) .add(MI.getOperand(2)) .add(MI.getOperand(3)); BuildMI(MBB, MI, DL, get(Hexagon::V6_vS32b_new_ai)) .add(MI.getOperand(0)) .addImm(0) .addReg(Hexagon::VTMP); MBB.erase(MI); return First.getInstrIterator(); case Hexagon::V6_vgathermhw_pseudo: First = BuildMI(MBB, MI, DL, get(Hexagon::V6_vgathermhw)) .add(MI.getOperand(1)) .add(MI.getOperand(2)) .add(MI.getOperand(3)); BuildMI(MBB, MI, DL, get(Hexagon::V6_vS32b_new_ai)) .add(MI.getOperand(0)) .addImm(0) .addReg(Hexagon::VTMP); MBB.erase(MI); return First.getInstrIterator(); case Hexagon::V6_vgathermhq_pseudo: First = BuildMI(MBB, MI, DL, get(Hexagon::V6_vgathermhq)) .add(MI.getOperand(1)) .add(MI.getOperand(2)) .add(MI.getOperand(3)) .add(MI.getOperand(4)); BuildMI(MBB, MI, DL, get(Hexagon::V6_vS32b_new_ai)) .add(MI.getOperand(0)) .addImm(0) .addReg(Hexagon::VTMP); MBB.erase(MI); return First.getInstrIterator(); case Hexagon::V6_vgathermwq_pseudo: First = BuildMI(MBB, MI, DL, get(Hexagon::V6_vgathermwq)) .add(MI.getOperand(1)) .add(MI.getOperand(2)) .add(MI.getOperand(3)) .add(MI.getOperand(4)); BuildMI(MBB, MI, DL, get(Hexagon::V6_vS32b_new_ai)) .add(MI.getOperand(0)) .addImm(0) .addReg(Hexagon::VTMP); MBB.erase(MI); return First.getInstrIterator(); case Hexagon::V6_vgathermhwq_pseudo: First = BuildMI(MBB, MI, DL, get(Hexagon::V6_vgathermhwq)) .add(MI.getOperand(1)) .add(MI.getOperand(2)) .add(MI.getOperand(3)) .add(MI.getOperand(4)); BuildMI(MBB, MI, DL, get(Hexagon::V6_vS32b_new_ai)) .add(MI.getOperand(0)) .addImm(0) .addReg(Hexagon::VTMP); MBB.erase(MI); return First.getInstrIterator(); } return MI.getIterator(); } // We indicate that we want to reverse the branch by // inserting the reversed branching opcode. bool HexagonInstrInfo::reverseBranchCondition( SmallVectorImpl &Cond) const { if (Cond.empty()) return true; assert(Cond[0].isImm() && "First entry in the cond vector not imm-val"); unsigned opcode = Cond[0].getImm(); //unsigned temp; assert(get(opcode).isBranch() && "Should be a branching condition."); if (isEndLoopN(opcode)) return true; unsigned NewOpcode = getInvertedPredicatedOpcode(opcode); Cond[0].setImm(NewOpcode); return false; } void HexagonInstrInfo::insertNoop(MachineBasicBlock &MBB, MachineBasicBlock::iterator MI) const { DebugLoc DL; BuildMI(MBB, MI, DL, get(Hexagon::A2_nop)); } bool HexagonInstrInfo::isPostIncrement(const MachineInstr &MI) const { return getAddrMode(MI) == HexagonII::PostInc; } // Returns true if an instruction is predicated irrespective of the predicate // sense. For example, all of the following will return true. // if (p0) R1 = add(R2, R3) // if (!p0) R1 = add(R2, R3) // if (p0.new) R1 = add(R2, R3) // if (!p0.new) R1 = add(R2, R3) // Note: New-value stores are not included here as in the current // implementation, we don't need to check their predicate sense. bool HexagonInstrInfo::isPredicated(const MachineInstr &MI) const { const uint64_t F = MI.getDesc().TSFlags; return (F >> HexagonII::PredicatedPos) & HexagonII::PredicatedMask; } bool HexagonInstrInfo::PredicateInstruction( MachineInstr &MI, ArrayRef Cond) const { if (Cond.empty() || isNewValueJump(Cond[0].getImm()) || isEndLoopN(Cond[0].getImm())) { LLVM_DEBUG(dbgs() << "\nCannot predicate:"; MI.dump();); return false; } int Opc = MI.getOpcode(); assert (isPredicable(MI) && "Expected predicable instruction"); bool invertJump = predOpcodeHasNot(Cond); // We have to predicate MI "in place", i.e. after this function returns, // MI will need to be transformed into a predicated form. To avoid com- // plicated manipulations with the operands (handling tied operands, // etc.), build a new temporary instruction, then overwrite MI with it. MachineBasicBlock &B = *MI.getParent(); DebugLoc DL = MI.getDebugLoc(); unsigned PredOpc = getCondOpcode(Opc, invertJump); MachineInstrBuilder T = BuildMI(B, MI, DL, get(PredOpc)); unsigned NOp = 0, NumOps = MI.getNumOperands(); while (NOp < NumOps) { MachineOperand &Op = MI.getOperand(NOp); if (!Op.isReg() || !Op.isDef() || Op.isImplicit()) break; T.add(Op); NOp++; } unsigned PredReg, PredRegPos, PredRegFlags; bool GotPredReg = getPredReg(Cond, PredReg, PredRegPos, PredRegFlags); (void)GotPredReg; assert(GotPredReg); T.addReg(PredReg, PredRegFlags); while (NOp < NumOps) T.add(MI.getOperand(NOp++)); MI.setDesc(get(PredOpc)); while (unsigned n = MI.getNumOperands()) MI.RemoveOperand(n-1); for (unsigned i = 0, n = T->getNumOperands(); i < n; ++i) MI.addOperand(T->getOperand(i)); MachineBasicBlock::instr_iterator TI = T->getIterator(); B.erase(TI); MachineRegisterInfo &MRI = B.getParent()->getRegInfo(); MRI.clearKillFlags(PredReg); return true; } bool HexagonInstrInfo::SubsumesPredicate(ArrayRef Pred1, ArrayRef Pred2) const { // TODO: Fix this return false; } bool HexagonInstrInfo::ClobbersPredicate(MachineInstr &MI, std::vector &Pred, bool SkipDead) const { const HexagonRegisterInfo &HRI = *Subtarget.getRegisterInfo(); for (unsigned oper = 0; oper < MI.getNumOperands(); ++oper) { MachineOperand MO = MI.getOperand(oper); if (MO.isReg()) { if (!MO.isDef()) continue; const TargetRegisterClass* RC = HRI.getMinimalPhysRegClass(MO.getReg()); if (RC == &Hexagon::PredRegsRegClass) { Pred.push_back(MO); return true; } continue; } else if (MO.isRegMask()) { for (unsigned PR : Hexagon::PredRegsRegClass) { if (!MI.modifiesRegister(PR, &HRI)) continue; Pred.push_back(MO); return true; } } } return false; } bool HexagonInstrInfo::isPredicable(const MachineInstr &MI) const { if (!MI.getDesc().isPredicable()) return false; if (MI.isCall() || isTailCall(MI)) { if (!Subtarget.usePredicatedCalls()) return false; } // HVX loads are not predicable on v60, but are on v62. if (!Subtarget.hasV62Ops()) { switch (MI.getOpcode()) { case Hexagon::V6_vL32b_ai: case Hexagon::V6_vL32b_pi: case Hexagon::V6_vL32b_ppu: case Hexagon::V6_vL32b_cur_ai: case Hexagon::V6_vL32b_cur_pi: case Hexagon::V6_vL32b_cur_ppu: case Hexagon::V6_vL32b_nt_ai: case Hexagon::V6_vL32b_nt_pi: case Hexagon::V6_vL32b_nt_ppu: case Hexagon::V6_vL32b_tmp_ai: case Hexagon::V6_vL32b_tmp_pi: case Hexagon::V6_vL32b_tmp_ppu: case Hexagon::V6_vL32b_nt_cur_ai: case Hexagon::V6_vL32b_nt_cur_pi: case Hexagon::V6_vL32b_nt_cur_ppu: case Hexagon::V6_vL32b_nt_tmp_ai: case Hexagon::V6_vL32b_nt_tmp_pi: case Hexagon::V6_vL32b_nt_tmp_ppu: return false; } } return true; } bool HexagonInstrInfo::isSchedulingBoundary(const MachineInstr &MI, const MachineBasicBlock *MBB, const MachineFunction &MF) const { // Debug info is never a scheduling boundary. It's necessary to be explicit // due to the special treatment of IT instructions below, otherwise a // dbg_value followed by an IT will result in the IT instruction being // considered a scheduling hazard, which is wrong. It should be the actual // instruction preceding the dbg_value instruction(s), just like it is // when debug info is not present. if (MI.isDebugInstr()) return false; // Throwing call is a boundary. if (MI.isCall()) { // Don't mess around with no return calls. if (doesNotReturn(MI)) return true; // If any of the block's successors is a landing pad, this could be a // throwing call. for (auto I : MBB->successors()) if (I->isEHPad()) return true; } // Terminators and labels can't be scheduled around. if (MI.getDesc().isTerminator() || MI.isPosition()) return true; // INLINEASM_BR can jump to another block if (MI.getOpcode() == TargetOpcode::INLINEASM_BR) return true; if (MI.isInlineAsm() && !ScheduleInlineAsm) return true; return false; } /// Measure the specified inline asm to determine an approximation of its /// length. /// Comments (which run till the next SeparatorString or newline) do not /// count as an instruction. /// Any other non-whitespace text is considered an instruction, with /// multiple instructions separated by SeparatorString or newlines. /// Variable-length instructions are not handled here; this function /// may be overloaded in the target code to do that. /// Hexagon counts the number of ##'s and adjust for that many /// constant exenders. unsigned HexagonInstrInfo::getInlineAsmLength(const char *Str, const MCAsmInfo &MAI, const TargetSubtargetInfo *STI) const { StringRef AStr(Str); // Count the number of instructions in the asm. bool atInsnStart = true; unsigned Length = 0; const unsigned MaxInstLength = MAI.getMaxInstLength(STI); for (; *Str; ++Str) { if (*Str == '\n' || strncmp(Str, MAI.getSeparatorString(), strlen(MAI.getSeparatorString())) == 0) atInsnStart = true; if (atInsnStart && !isSpace(static_cast(*Str))) { Length += MaxInstLength; atInsnStart = false; } if (atInsnStart && strncmp(Str, MAI.getCommentString().data(), MAI.getCommentString().size()) == 0) atInsnStart = false; } // Add to size number of constant extenders seen * 4. StringRef Occ("##"); Length += AStr.count(Occ)*4; return Length; } ScheduleHazardRecognizer* HexagonInstrInfo::CreateTargetPostRAHazardRecognizer( const InstrItineraryData *II, const ScheduleDAG *DAG) const { if (UseDFAHazardRec) return new HexagonHazardRecognizer(II, this, Subtarget); return TargetInstrInfo::CreateTargetPostRAHazardRecognizer(II, DAG); } /// For a comparison instruction, return the source registers in /// \p SrcReg and \p SrcReg2 if having two register operands, and the value it /// compares against in CmpValue. Return true if the comparison instruction /// can be analyzed. bool HexagonInstrInfo::analyzeCompare(const MachineInstr &MI, Register &SrcReg, Register &SrcReg2, int &Mask, int &Value) const { unsigned Opc = MI.getOpcode(); // Set mask and the first source register. switch (Opc) { case Hexagon::C2_cmpeq: case Hexagon::C2_cmpeqp: case Hexagon::C2_cmpgt: case Hexagon::C2_cmpgtp: case Hexagon::C2_cmpgtu: case Hexagon::C2_cmpgtup: case Hexagon::C4_cmpneq: case Hexagon::C4_cmplte: case Hexagon::C4_cmplteu: case Hexagon::C2_cmpeqi: case Hexagon::C2_cmpgti: case Hexagon::C2_cmpgtui: case Hexagon::C4_cmpneqi: case Hexagon::C4_cmplteui: case Hexagon::C4_cmpltei: SrcReg = MI.getOperand(1).getReg(); Mask = ~0; break; case Hexagon::A4_cmpbeq: case Hexagon::A4_cmpbgt: case Hexagon::A4_cmpbgtu: case Hexagon::A4_cmpbeqi: case Hexagon::A4_cmpbgti: case Hexagon::A4_cmpbgtui: SrcReg = MI.getOperand(1).getReg(); Mask = 0xFF; break; case Hexagon::A4_cmpheq: case Hexagon::A4_cmphgt: case Hexagon::A4_cmphgtu: case Hexagon::A4_cmpheqi: case Hexagon::A4_cmphgti: case Hexagon::A4_cmphgtui: SrcReg = MI.getOperand(1).getReg(); Mask = 0xFFFF; break; } // Set the value/second source register. switch (Opc) { case Hexagon::C2_cmpeq: case Hexagon::C2_cmpeqp: case Hexagon::C2_cmpgt: case Hexagon::C2_cmpgtp: case Hexagon::C2_cmpgtu: case Hexagon::C2_cmpgtup: case Hexagon::A4_cmpbeq: case Hexagon::A4_cmpbgt: case Hexagon::A4_cmpbgtu: case Hexagon::A4_cmpheq: case Hexagon::A4_cmphgt: case Hexagon::A4_cmphgtu: case Hexagon::C4_cmpneq: case Hexagon::C4_cmplte: case Hexagon::C4_cmplteu: SrcReg2 = MI.getOperand(2).getReg(); return true; case Hexagon::C2_cmpeqi: case Hexagon::C2_cmpgtui: case Hexagon::C2_cmpgti: case Hexagon::C4_cmpneqi: case Hexagon::C4_cmplteui: case Hexagon::C4_cmpltei: case Hexagon::A4_cmpbeqi: case Hexagon::A4_cmpbgti: case Hexagon::A4_cmpbgtui: case Hexagon::A4_cmpheqi: case Hexagon::A4_cmphgti: case Hexagon::A4_cmphgtui: { SrcReg2 = 0; const MachineOperand &Op2 = MI.getOperand(2); if (!Op2.isImm()) return false; Value = MI.getOperand(2).getImm(); return true; } } return false; } unsigned HexagonInstrInfo::getInstrLatency(const InstrItineraryData *ItinData, const MachineInstr &MI, unsigned *PredCost) const { return getInstrTimingClassLatency(ItinData, MI); } DFAPacketizer *HexagonInstrInfo::CreateTargetScheduleState( const TargetSubtargetInfo &STI) const { const InstrItineraryData *II = STI.getInstrItineraryData(); return static_cast(STI).createDFAPacketizer(II); } // Inspired by this pair: // %r13 = L2_loadri_io %r29, 136; mem:LD4[FixedStack0] // S2_storeri_io %r29, 132, killed %r1; flags: mem:ST4[FixedStack1] // Currently AA considers the addresses in these instructions to be aliasing. bool HexagonInstrInfo::areMemAccessesTriviallyDisjoint( const MachineInstr &MIa, const MachineInstr &MIb) const { if (MIa.hasUnmodeledSideEffects() || MIb.hasUnmodeledSideEffects() || MIa.hasOrderedMemoryRef() || MIb.hasOrderedMemoryRef()) return false; // Instructions that are pure loads, not loads and stores like memops are not // dependent. if (MIa.mayLoad() && !isMemOp(MIa) && MIb.mayLoad() && !isMemOp(MIb)) return true; // Get the base register in MIa. unsigned BasePosA, OffsetPosA; if (!getBaseAndOffsetPosition(MIa, BasePosA, OffsetPosA)) return false; const MachineOperand &BaseA = MIa.getOperand(BasePosA); Register BaseRegA = BaseA.getReg(); unsigned BaseSubA = BaseA.getSubReg(); // Get the base register in MIb. unsigned BasePosB, OffsetPosB; if (!getBaseAndOffsetPosition(MIb, BasePosB, OffsetPosB)) return false; const MachineOperand &BaseB = MIb.getOperand(BasePosB); Register BaseRegB = BaseB.getReg(); unsigned BaseSubB = BaseB.getSubReg(); if (BaseRegA != BaseRegB || BaseSubA != BaseSubB) return false; // Get the access sizes. unsigned SizeA = getMemAccessSize(MIa); unsigned SizeB = getMemAccessSize(MIb); // Get the offsets. Handle immediates only for now. const MachineOperand &OffA = MIa.getOperand(OffsetPosA); const MachineOperand &OffB = MIb.getOperand(OffsetPosB); if (!MIa.getOperand(OffsetPosA).isImm() || !MIb.getOperand(OffsetPosB).isImm()) return false; int OffsetA = isPostIncrement(MIa) ? 0 : OffA.getImm(); int OffsetB = isPostIncrement(MIb) ? 0 : OffB.getImm(); // This is a mem access with the same base register and known offsets from it. // Reason about it. if (OffsetA > OffsetB) { uint64_t OffDiff = (uint64_t)((int64_t)OffsetA - (int64_t)OffsetB); return SizeB <= OffDiff; } if (OffsetA < OffsetB) { uint64_t OffDiff = (uint64_t)((int64_t)OffsetB - (int64_t)OffsetA); return SizeA <= OffDiff; } return false; } /// If the instruction is an increment of a constant value, return the amount. bool HexagonInstrInfo::getIncrementValue(const MachineInstr &MI, int &Value) const { if (isPostIncrement(MI)) { unsigned BasePos = 0, OffsetPos = 0; if (!getBaseAndOffsetPosition(MI, BasePos, OffsetPos)) return false; const MachineOperand &OffsetOp = MI.getOperand(OffsetPos); if (OffsetOp.isImm()) { Value = OffsetOp.getImm(); return true; } } else if (MI.getOpcode() == Hexagon::A2_addi) { const MachineOperand &AddOp = MI.getOperand(2); if (AddOp.isImm()) { Value = AddOp.getImm(); return true; } } return false; } std::pair HexagonInstrInfo::decomposeMachineOperandsTargetFlags(unsigned TF) const { return std::make_pair(TF & ~HexagonII::MO_Bitmasks, TF & HexagonII::MO_Bitmasks); } ArrayRef> HexagonInstrInfo::getSerializableDirectMachineOperandTargetFlags() const { using namespace HexagonII; static const std::pair Flags[] = { {MO_PCREL, "hexagon-pcrel"}, {MO_GOT, "hexagon-got"}, {MO_LO16, "hexagon-lo16"}, {MO_HI16, "hexagon-hi16"}, {MO_GPREL, "hexagon-gprel"}, {MO_GDGOT, "hexagon-gdgot"}, {MO_GDPLT, "hexagon-gdplt"}, {MO_IE, "hexagon-ie"}, {MO_IEGOT, "hexagon-iegot"}, {MO_TPREL, "hexagon-tprel"} }; return makeArrayRef(Flags); } ArrayRef> HexagonInstrInfo::getSerializableBitmaskMachineOperandTargetFlags() const { using namespace HexagonII; static const std::pair Flags[] = { {HMOTF_ConstExtended, "hexagon-ext"} }; return makeArrayRef(Flags); } unsigned HexagonInstrInfo::createVR(MachineFunction *MF, MVT VT) const { MachineRegisterInfo &MRI = MF->getRegInfo(); const TargetRegisterClass *TRC; if (VT == MVT::i1) { TRC = &Hexagon::PredRegsRegClass; } else if (VT == MVT::i32 || VT == MVT::f32) { TRC = &Hexagon::IntRegsRegClass; } else if (VT == MVT::i64 || VT == MVT::f64) { TRC = &Hexagon::DoubleRegsRegClass; } else { llvm_unreachable("Cannot handle this register class"); } Register NewReg = MRI.createVirtualRegister(TRC); return NewReg; } bool HexagonInstrInfo::isAbsoluteSet(const MachineInstr &MI) const { return (getAddrMode(MI) == HexagonII::AbsoluteSet); } bool HexagonInstrInfo::isAccumulator(const MachineInstr &MI) const { const uint64_t F = MI.getDesc().TSFlags; return((F >> HexagonII::AccumulatorPos) & HexagonII::AccumulatorMask); } bool HexagonInstrInfo::isBaseImmOffset(const MachineInstr &MI) const { return getAddrMode(MI) == HexagonII::BaseImmOffset; } bool HexagonInstrInfo::isComplex(const MachineInstr &MI) const { return !isTC1(MI) && !isTC2Early(MI) && !MI.getDesc().mayLoad() && !MI.getDesc().mayStore() && MI.getDesc().getOpcode() != Hexagon::S2_allocframe && MI.getDesc().getOpcode() != Hexagon::L2_deallocframe && !isMemOp(MI) && !MI.isBranch() && !MI.isReturn() && !MI.isCall(); } // Return true if the instruction is a compund branch instruction. bool HexagonInstrInfo::isCompoundBranchInstr(const MachineInstr &MI) const { return getType(MI) == HexagonII::TypeCJ && MI.isBranch(); } // TODO: In order to have isExtendable for fpimm/f32Ext, we need to handle // isFPImm and later getFPImm as well. bool HexagonInstrInfo::isConstExtended(const MachineInstr &MI) const { const uint64_t F = MI.getDesc().TSFlags; unsigned isExtended = (F >> HexagonII::ExtendedPos) & HexagonII::ExtendedMask; if (isExtended) // Instruction must be extended. return true; unsigned isExtendable = (F >> HexagonII::ExtendablePos) & HexagonII::ExtendableMask; if (!isExtendable) return false; if (MI.isCall()) return false; short ExtOpNum = getCExtOpNum(MI); const MachineOperand &MO = MI.getOperand(ExtOpNum); // Use MO operand flags to determine if MO // has the HMOTF_ConstExtended flag set. if (MO.getTargetFlags() & HexagonII::HMOTF_ConstExtended) return true; // If this is a Machine BB address we are talking about, and it is // not marked as extended, say so. if (MO.isMBB()) return false; // We could be using an instruction with an extendable immediate and shoehorn // a global address into it. If it is a global address it will be constant // extended. We do this for COMBINE. if (MO.isGlobal() || MO.isSymbol() || MO.isBlockAddress() || MO.isJTI() || MO.isCPI() || MO.isFPImm()) return true; // If the extendable operand is not 'Immediate' type, the instruction should // have 'isExtended' flag set. assert(MO.isImm() && "Extendable operand must be Immediate type"); int MinValue = getMinValue(MI); int MaxValue = getMaxValue(MI); int ImmValue = MO.getImm(); return (ImmValue < MinValue || ImmValue > MaxValue); } bool HexagonInstrInfo::isDeallocRet(const MachineInstr &MI) const { switch (MI.getOpcode()) { case Hexagon::L4_return: case Hexagon::L4_return_t: case Hexagon::L4_return_f: case Hexagon::L4_return_tnew_pnt: case Hexagon::L4_return_fnew_pnt: case Hexagon::L4_return_tnew_pt: case Hexagon::L4_return_fnew_pt: return true; } return false; } // Return true when ConsMI uses a register defined by ProdMI. bool HexagonInstrInfo::isDependent(const MachineInstr &ProdMI, const MachineInstr &ConsMI) const { if (!ProdMI.getDesc().getNumDefs()) return false; const HexagonRegisterInfo &HRI = *Subtarget.getRegisterInfo(); SmallVector DefsA; SmallVector DefsB; SmallVector UsesA; SmallVector UsesB; parseOperands(ProdMI, DefsA, UsesA); parseOperands(ConsMI, DefsB, UsesB); for (auto &RegA : DefsA) for (auto &RegB : UsesB) { // True data dependency. if (RegA == RegB) return true; if (Register::isPhysicalRegister(RegA)) for (MCSubRegIterator SubRegs(RegA, &HRI); SubRegs.isValid(); ++SubRegs) if (RegB == *SubRegs) return true; if (Register::isPhysicalRegister(RegB)) for (MCSubRegIterator SubRegs(RegB, &HRI); SubRegs.isValid(); ++SubRegs) if (RegA == *SubRegs) return true; } return false; } // Returns true if the instruction is alread a .cur. bool HexagonInstrInfo::isDotCurInst(const MachineInstr &MI) const { switch (MI.getOpcode()) { case Hexagon::V6_vL32b_cur_pi: case Hexagon::V6_vL32b_cur_ai: return true; } return false; } // Returns true, if any one of the operands is a dot new // insn, whether it is predicated dot new or register dot new. bool HexagonInstrInfo::isDotNewInst(const MachineInstr &MI) const { if (isNewValueInst(MI) || (isPredicated(MI) && isPredicatedNew(MI))) return true; return false; } /// Symmetrical. See if these two instructions are fit for duplex pair. bool HexagonInstrInfo::isDuplexPair(const MachineInstr &MIa, const MachineInstr &MIb) const { HexagonII::SubInstructionGroup MIaG = getDuplexCandidateGroup(MIa); HexagonII::SubInstructionGroup MIbG = getDuplexCandidateGroup(MIb); return (isDuplexPairMatch(MIaG, MIbG) || isDuplexPairMatch(MIbG, MIaG)); } bool HexagonInstrInfo::isEarlySourceInstr(const MachineInstr &MI) const { if (MI.mayLoadOrStore() || MI.isCompare()) return true; // Multiply unsigned SchedClass = MI.getDesc().getSchedClass(); return is_TC4x(SchedClass) || is_TC3x(SchedClass); } bool HexagonInstrInfo::isEndLoopN(unsigned Opcode) const { return (Opcode == Hexagon::ENDLOOP0 || Opcode == Hexagon::ENDLOOP1); } bool HexagonInstrInfo::isExpr(unsigned OpType) const { switch(OpType) { case MachineOperand::MO_MachineBasicBlock: case MachineOperand::MO_GlobalAddress: case MachineOperand::MO_ExternalSymbol: case MachineOperand::MO_JumpTableIndex: case MachineOperand::MO_ConstantPoolIndex: case MachineOperand::MO_BlockAddress: return true; default: return false; } } bool HexagonInstrInfo::isExtendable(const MachineInstr &MI) const { const MCInstrDesc &MID = MI.getDesc(); const uint64_t F = MID.TSFlags; if ((F >> HexagonII::ExtendablePos) & HexagonII::ExtendableMask) return true; // TODO: This is largely obsolete now. Will need to be removed // in consecutive patches. switch (MI.getOpcode()) { // PS_fi and PS_fia remain special cases. case Hexagon::PS_fi: case Hexagon::PS_fia: return true; default: return false; } return false; } // This returns true in two cases: // - The OP code itself indicates that this is an extended instruction. // - One of MOs has been marked with HMOTF_ConstExtended flag. bool HexagonInstrInfo::isExtended(const MachineInstr &MI) const { // First check if this is permanently extended op code. const uint64_t F = MI.getDesc().TSFlags; if ((F >> HexagonII::ExtendedPos) & HexagonII::ExtendedMask) return true; // Use MO operand flags to determine if one of MI's operands // has HMOTF_ConstExtended flag set. for (const MachineOperand &MO : MI.operands()) if (MO.getTargetFlags() & HexagonII::HMOTF_ConstExtended) return true; return false; } bool HexagonInstrInfo::isFloat(const MachineInstr &MI) const { unsigned Opcode = MI.getOpcode(); const uint64_t F = get(Opcode).TSFlags; return (F >> HexagonII::FPPos) & HexagonII::FPMask; } // No V60 HVX VMEM with A_INDIRECT. bool HexagonInstrInfo::isHVXMemWithAIndirect(const MachineInstr &I, const MachineInstr &J) const { if (!isHVXVec(I)) return false; if (!I.mayLoad() && !I.mayStore()) return false; return J.isIndirectBranch() || isIndirectCall(J) || isIndirectL4Return(J); } bool HexagonInstrInfo::isIndirectCall(const MachineInstr &MI) const { switch (MI.getOpcode()) { case Hexagon::J2_callr: case Hexagon::J2_callrf: case Hexagon::J2_callrt: case Hexagon::PS_call_nr: return true; } return false; } bool HexagonInstrInfo::isIndirectL4Return(const MachineInstr &MI) const { switch (MI.getOpcode()) { case Hexagon::L4_return: case Hexagon::L4_return_t: case Hexagon::L4_return_f: case Hexagon::L4_return_fnew_pnt: case Hexagon::L4_return_fnew_pt: case Hexagon::L4_return_tnew_pnt: case Hexagon::L4_return_tnew_pt: return true; } return false; } bool HexagonInstrInfo::isJumpR(const MachineInstr &MI) const { switch (MI.getOpcode()) { case Hexagon::J2_jumpr: case Hexagon::J2_jumprt: case Hexagon::J2_jumprf: case Hexagon::J2_jumprtnewpt: case Hexagon::J2_jumprfnewpt: case Hexagon::J2_jumprtnew: case Hexagon::J2_jumprfnew: return true; } return false; } // Return true if a given MI can accommodate given offset. // Use abs estimate as oppose to the exact number. // TODO: This will need to be changed to use MC level // definition of instruction extendable field size. bool HexagonInstrInfo::isJumpWithinBranchRange(const MachineInstr &MI, unsigned offset) const { // This selection of jump instructions matches to that what // analyzeBranch can parse, plus NVJ. if (isNewValueJump(MI)) // r9:2 return isInt<11>(offset); switch (MI.getOpcode()) { // Still missing Jump to address condition on register value. default: return false; case Hexagon::J2_jump: // bits<24> dst; // r22:2 case Hexagon::J2_call: case Hexagon::PS_call_nr: return isInt<24>(offset); case Hexagon::J2_jumpt: //bits<17> dst; // r15:2 case Hexagon::J2_jumpf: case Hexagon::J2_jumptnew: case Hexagon::J2_jumptnewpt: case Hexagon::J2_jumpfnew: case Hexagon::J2_jumpfnewpt: case Hexagon::J2_callt: case Hexagon::J2_callf: return isInt<17>(offset); case Hexagon::J2_loop0i: case Hexagon::J2_loop0iext: case Hexagon::J2_loop0r: case Hexagon::J2_loop0rext: case Hexagon::J2_loop1i: case Hexagon::J2_loop1iext: case Hexagon::J2_loop1r: case Hexagon::J2_loop1rext: return isInt<9>(offset); // TODO: Add all the compound branches here. Can we do this in Relation model? case Hexagon::J4_cmpeqi_tp0_jump_nt: case Hexagon::J4_cmpeqi_tp1_jump_nt: case Hexagon::J4_cmpeqn1_tp0_jump_nt: case Hexagon::J4_cmpeqn1_tp1_jump_nt: return isInt<11>(offset); } } bool HexagonInstrInfo::isLateInstrFeedsEarlyInstr(const MachineInstr &LRMI, const MachineInstr &ESMI) const { bool isLate = isLateResultInstr(LRMI); bool isEarly = isEarlySourceInstr(ESMI); LLVM_DEBUG(dbgs() << "V60" << (isLate ? "-LR " : " -- ")); LLVM_DEBUG(LRMI.dump()); LLVM_DEBUG(dbgs() << "V60" << (isEarly ? "-ES " : " -- ")); LLVM_DEBUG(ESMI.dump()); if (isLate && isEarly) { LLVM_DEBUG(dbgs() << "++Is Late Result feeding Early Source\n"); return true; } return false; } bool HexagonInstrInfo::isLateResultInstr(const MachineInstr &MI) const { switch (MI.getOpcode()) { case TargetOpcode::EXTRACT_SUBREG: case TargetOpcode::INSERT_SUBREG: case TargetOpcode::SUBREG_TO_REG: case TargetOpcode::REG_SEQUENCE: case TargetOpcode::IMPLICIT_DEF: case TargetOpcode::COPY: case TargetOpcode::INLINEASM: case TargetOpcode::PHI: return false; default: break; } unsigned SchedClass = MI.getDesc().getSchedClass(); return !is_TC1(SchedClass); } bool HexagonInstrInfo::isLateSourceInstr(const MachineInstr &MI) const { // Instructions with iclass A_CVI_VX and attribute A_CVI_LATE uses a multiply // resource, but all operands can be received late like an ALU instruction. return getType(MI) == HexagonII::TypeCVI_VX_LATE; } bool HexagonInstrInfo::isLoopN(const MachineInstr &MI) const { unsigned Opcode = MI.getOpcode(); return Opcode == Hexagon::J2_loop0i || Opcode == Hexagon::J2_loop0r || Opcode == Hexagon::J2_loop0iext || Opcode == Hexagon::J2_loop0rext || Opcode == Hexagon::J2_loop1i || Opcode == Hexagon::J2_loop1r || Opcode == Hexagon::J2_loop1iext || Opcode == Hexagon::J2_loop1rext; } bool HexagonInstrInfo::isMemOp(const MachineInstr &MI) const { switch (MI.getOpcode()) { default: return false; case Hexagon::L4_iadd_memopw_io: case Hexagon::L4_isub_memopw_io: case Hexagon::L4_add_memopw_io: case Hexagon::L4_sub_memopw_io: case Hexagon::L4_and_memopw_io: case Hexagon::L4_or_memopw_io: case Hexagon::L4_iadd_memoph_io: case Hexagon::L4_isub_memoph_io: case Hexagon::L4_add_memoph_io: case Hexagon::L4_sub_memoph_io: case Hexagon::L4_and_memoph_io: case Hexagon::L4_or_memoph_io: case Hexagon::L4_iadd_memopb_io: case Hexagon::L4_isub_memopb_io: case Hexagon::L4_add_memopb_io: case Hexagon::L4_sub_memopb_io: case Hexagon::L4_and_memopb_io: case Hexagon::L4_or_memopb_io: case Hexagon::L4_ior_memopb_io: case Hexagon::L4_ior_memoph_io: case Hexagon::L4_ior_memopw_io: case Hexagon::L4_iand_memopb_io: case Hexagon::L4_iand_memoph_io: case Hexagon::L4_iand_memopw_io: return true; } return false; } bool HexagonInstrInfo::isNewValue(const MachineInstr &MI) const { const uint64_t F = MI.getDesc().TSFlags; return (F >> HexagonII::NewValuePos) & HexagonII::NewValueMask; } bool HexagonInstrInfo::isNewValue(unsigned Opcode) const { const uint64_t F = get(Opcode).TSFlags; return (F >> HexagonII::NewValuePos) & HexagonII::NewValueMask; } bool HexagonInstrInfo::isNewValueInst(const MachineInstr &MI) const { return isNewValueJump(MI) || isNewValueStore(MI); } bool HexagonInstrInfo::isNewValueJump(const MachineInstr &MI) const { return isNewValue(MI) && MI.isBranch(); } bool HexagonInstrInfo::isNewValueJump(unsigned Opcode) const { return isNewValue(Opcode) && get(Opcode).isBranch() && isPredicated(Opcode); } bool HexagonInstrInfo::isNewValueStore(const MachineInstr &MI) const { const uint64_t F = MI.getDesc().TSFlags; return (F >> HexagonII::NVStorePos) & HexagonII::NVStoreMask; } bool HexagonInstrInfo::isNewValueStore(unsigned Opcode) const { const uint64_t F = get(Opcode).TSFlags; return (F >> HexagonII::NVStorePos) & HexagonII::NVStoreMask; } // Returns true if a particular operand is extendable for an instruction. bool HexagonInstrInfo::isOperandExtended(const MachineInstr &MI, unsigned OperandNum) const { const uint64_t F = MI.getDesc().TSFlags; return ((F >> HexagonII::ExtendableOpPos) & HexagonII::ExtendableOpMask) == OperandNum; } bool HexagonInstrInfo::isPredicatedNew(const MachineInstr &MI) const { const uint64_t F = MI.getDesc().TSFlags; assert(isPredicated(MI)); return (F >> HexagonII::PredicatedNewPos) & HexagonII::PredicatedNewMask; } bool HexagonInstrInfo::isPredicatedNew(unsigned Opcode) const { const uint64_t F = get(Opcode).TSFlags; assert(isPredicated(Opcode)); return (F >> HexagonII::PredicatedNewPos) & HexagonII::PredicatedNewMask; } bool HexagonInstrInfo::isPredicatedTrue(const MachineInstr &MI) const { const uint64_t F = MI.getDesc().TSFlags; return !((F >> HexagonII::PredicatedFalsePos) & HexagonII::PredicatedFalseMask); } bool HexagonInstrInfo::isPredicatedTrue(unsigned Opcode) const { const uint64_t F = get(Opcode).TSFlags; // Make sure that the instruction is predicated. assert((F>> HexagonII::PredicatedPos) & HexagonII::PredicatedMask); return !((F >> HexagonII::PredicatedFalsePos) & HexagonII::PredicatedFalseMask); } bool HexagonInstrInfo::isPredicated(unsigned Opcode) const { const uint64_t F = get(Opcode).TSFlags; return (F >> HexagonII::PredicatedPos) & HexagonII::PredicatedMask; } bool HexagonInstrInfo::isPredicateLate(unsigned Opcode) const { const uint64_t F = get(Opcode).TSFlags; return (F >> HexagonII::PredicateLatePos) & HexagonII::PredicateLateMask; } bool HexagonInstrInfo::isPredictedTaken(unsigned Opcode) const { const uint64_t F = get(Opcode).TSFlags; assert(get(Opcode).isBranch() && (isPredicatedNew(Opcode) || isNewValue(Opcode))); return (F >> HexagonII::TakenPos) & HexagonII::TakenMask; } bool HexagonInstrInfo::isSaveCalleeSavedRegsCall(const MachineInstr &MI) const { return MI.getOpcode() == Hexagon::SAVE_REGISTERS_CALL_V4 || MI.getOpcode() == Hexagon::SAVE_REGISTERS_CALL_V4_EXT || MI.getOpcode() == Hexagon::SAVE_REGISTERS_CALL_V4_PIC || MI.getOpcode() == Hexagon::SAVE_REGISTERS_CALL_V4_EXT_PIC; } bool HexagonInstrInfo::isSignExtendingLoad(const MachineInstr &MI) const { switch (MI.getOpcode()) { // Byte case Hexagon::L2_loadrb_io: case Hexagon::L4_loadrb_ur: case Hexagon::L4_loadrb_ap: case Hexagon::L2_loadrb_pr: case Hexagon::L2_loadrb_pbr: case Hexagon::L2_loadrb_pi: case Hexagon::L2_loadrb_pci: case Hexagon::L2_loadrb_pcr: case Hexagon::L2_loadbsw2_io: case Hexagon::L4_loadbsw2_ur: case Hexagon::L4_loadbsw2_ap: case Hexagon::L2_loadbsw2_pr: case Hexagon::L2_loadbsw2_pbr: case Hexagon::L2_loadbsw2_pi: case Hexagon::L2_loadbsw2_pci: case Hexagon::L2_loadbsw2_pcr: case Hexagon::L2_loadbsw4_io: case Hexagon::L4_loadbsw4_ur: case Hexagon::L4_loadbsw4_ap: case Hexagon::L2_loadbsw4_pr: case Hexagon::L2_loadbsw4_pbr: case Hexagon::L2_loadbsw4_pi: case Hexagon::L2_loadbsw4_pci: case Hexagon::L2_loadbsw4_pcr: case Hexagon::L4_loadrb_rr: case Hexagon::L2_ploadrbt_io: case Hexagon::L2_ploadrbt_pi: case Hexagon::L2_ploadrbf_io: case Hexagon::L2_ploadrbf_pi: case Hexagon::L2_ploadrbtnew_io: case Hexagon::L2_ploadrbfnew_io: case Hexagon::L4_ploadrbt_rr: case Hexagon::L4_ploadrbf_rr: case Hexagon::L4_ploadrbtnew_rr: case Hexagon::L4_ploadrbfnew_rr: case Hexagon::L2_ploadrbtnew_pi: case Hexagon::L2_ploadrbfnew_pi: case Hexagon::L4_ploadrbt_abs: case Hexagon::L4_ploadrbf_abs: case Hexagon::L4_ploadrbtnew_abs: case Hexagon::L4_ploadrbfnew_abs: case Hexagon::L2_loadrbgp: // Half case Hexagon::L2_loadrh_io: case Hexagon::L4_loadrh_ur: case Hexagon::L4_loadrh_ap: case Hexagon::L2_loadrh_pr: case Hexagon::L2_loadrh_pbr: case Hexagon::L2_loadrh_pi: case Hexagon::L2_loadrh_pci: case Hexagon::L2_loadrh_pcr: case Hexagon::L4_loadrh_rr: case Hexagon::L2_ploadrht_io: case Hexagon::L2_ploadrht_pi: case Hexagon::L2_ploadrhf_io: case Hexagon::L2_ploadrhf_pi: case Hexagon::L2_ploadrhtnew_io: case Hexagon::L2_ploadrhfnew_io: case Hexagon::L4_ploadrht_rr: case Hexagon::L4_ploadrhf_rr: case Hexagon::L4_ploadrhtnew_rr: case Hexagon::L4_ploadrhfnew_rr: case Hexagon::L2_ploadrhtnew_pi: case Hexagon::L2_ploadrhfnew_pi: case Hexagon::L4_ploadrht_abs: case Hexagon::L4_ploadrhf_abs: case Hexagon::L4_ploadrhtnew_abs: case Hexagon::L4_ploadrhfnew_abs: case Hexagon::L2_loadrhgp: return true; default: return false; } } bool HexagonInstrInfo::isSolo(const MachineInstr &MI) const { const uint64_t F = MI.getDesc().TSFlags; return (F >> HexagonII::SoloPos) & HexagonII::SoloMask; } bool HexagonInstrInfo::isSpillPredRegOp(const MachineInstr &MI) const { switch (MI.getOpcode()) { case Hexagon::STriw_pred: case Hexagon::LDriw_pred: return true; default: return false; } } bool HexagonInstrInfo::isTailCall(const MachineInstr &MI) const { if (!MI.isBranch()) return false; for (auto &Op : MI.operands()) if (Op.isGlobal() || Op.isSymbol()) return true; return false; } // Returns true when SU has a timing class TC1. bool HexagonInstrInfo::isTC1(const MachineInstr &MI) const { unsigned SchedClass = MI.getDesc().getSchedClass(); return is_TC1(SchedClass); } bool HexagonInstrInfo::isTC2(const MachineInstr &MI) const { unsigned SchedClass = MI.getDesc().getSchedClass(); return is_TC2(SchedClass); } bool HexagonInstrInfo::isTC2Early(const MachineInstr &MI) const { unsigned SchedClass = MI.getDesc().getSchedClass(); return is_TC2early(SchedClass); } bool HexagonInstrInfo::isTC4x(const MachineInstr &MI) const { unsigned SchedClass = MI.getDesc().getSchedClass(); return is_TC4x(SchedClass); } // Schedule this ASAP. bool HexagonInstrInfo::isToBeScheduledASAP(const MachineInstr &MI1, const MachineInstr &MI2) const { if (mayBeCurLoad(MI1)) { // if (result of SU is used in Next) return true; Register DstReg = MI1.getOperand(0).getReg(); int N = MI2.getNumOperands(); for (int I = 0; I < N; I++) if (MI2.getOperand(I).isReg() && DstReg == MI2.getOperand(I).getReg()) return true; } if (mayBeNewStore(MI2)) if (MI2.getOpcode() == Hexagon::V6_vS32b_pi) if (MI1.getOperand(0).isReg() && MI2.getOperand(3).isReg() && MI1.getOperand(0).getReg() == MI2.getOperand(3).getReg()) return true; return false; } bool HexagonInstrInfo::isHVXVec(const MachineInstr &MI) const { const uint64_t V = getType(MI); return HexagonII::TypeCVI_FIRST <= V && V <= HexagonII::TypeCVI_LAST; } // Check if the Offset is a valid auto-inc imm by Load/Store Type. bool HexagonInstrInfo::isValidAutoIncImm(const EVT VT, int Offset) const { int Size = VT.getSizeInBits() / 8; if (Offset % Size != 0) return false; int Count = Offset / Size; switch (VT.getSimpleVT().SimpleTy) { // For scalars the auto-inc is s4 case MVT::i8: case MVT::i16: case MVT::i32: case MVT::i64: case MVT::f32: case MVT::f64: case MVT::v2i16: case MVT::v2i32: case MVT::v4i8: case MVT::v4i16: case MVT::v8i8: return isInt<4>(Count); // For HVX vectors the auto-inc is s3 case MVT::v64i8: case MVT::v32i16: case MVT::v16i32: case MVT::v8i64: case MVT::v128i8: case MVT::v64i16: case MVT::v32i32: case MVT::v16i64: return isInt<3>(Count); default: break; } llvm_unreachable("Not an valid type!"); } bool HexagonInstrInfo::isValidOffset(unsigned Opcode, int Offset, const TargetRegisterInfo *TRI, bool Extend) const { // This function is to check whether the "Offset" is in the correct range of // the given "Opcode". If "Offset" is not in the correct range, "A2_addi" is // inserted to calculate the final address. Due to this reason, the function // assumes that the "Offset" has correct alignment. // We used to assert if the offset was not properly aligned, however, // there are cases where a misaligned pointer recast can cause this // problem, and we need to allow for it. The front end warns of such // misaligns with respect to load size. switch (Opcode) { case Hexagon::PS_vstorerq_ai: case Hexagon::PS_vstorerv_ai: case Hexagon::PS_vstorerw_ai: case Hexagon::PS_vstorerw_nt_ai: case Hexagon::PS_vloadrq_ai: case Hexagon::PS_vloadrv_ai: case Hexagon::PS_vloadrw_ai: case Hexagon::PS_vloadrw_nt_ai: case Hexagon::V6_vL32b_ai: case Hexagon::V6_vS32b_ai: case Hexagon::V6_vS32b_qpred_ai: case Hexagon::V6_vS32b_nqpred_ai: case Hexagon::V6_vL32b_nt_ai: case Hexagon::V6_vS32b_nt_ai: case Hexagon::V6_vL32Ub_ai: case Hexagon::V6_vS32Ub_ai: { unsigned VectorSize = TRI->getSpillSize(Hexagon::HvxVRRegClass); assert(isPowerOf2_32(VectorSize)); if (Offset & (VectorSize-1)) return false; return isInt<4>(Offset >> Log2_32(VectorSize)); } case Hexagon::J2_loop0i: case Hexagon::J2_loop1i: return isUInt<10>(Offset); case Hexagon::S4_storeirb_io: case Hexagon::S4_storeirbt_io: case Hexagon::S4_storeirbf_io: return isUInt<6>(Offset); case Hexagon::S4_storeirh_io: case Hexagon::S4_storeirht_io: case Hexagon::S4_storeirhf_io: return isShiftedUInt<6,1>(Offset); case Hexagon::S4_storeiri_io: case Hexagon::S4_storeirit_io: case Hexagon::S4_storeirif_io: return isShiftedUInt<6,2>(Offset); } if (Extend) return true; switch (Opcode) { case Hexagon::L2_loadri_io: case Hexagon::S2_storeri_io: return (Offset >= Hexagon_MEMW_OFFSET_MIN) && (Offset <= Hexagon_MEMW_OFFSET_MAX); case Hexagon::L2_loadrd_io: case Hexagon::S2_storerd_io: return (Offset >= Hexagon_MEMD_OFFSET_MIN) && (Offset <= Hexagon_MEMD_OFFSET_MAX); case Hexagon::L2_loadrh_io: case Hexagon::L2_loadruh_io: case Hexagon::S2_storerh_io: case Hexagon::S2_storerf_io: return (Offset >= Hexagon_MEMH_OFFSET_MIN) && (Offset <= Hexagon_MEMH_OFFSET_MAX); case Hexagon::L2_loadrb_io: case Hexagon::L2_loadrub_io: case Hexagon::S2_storerb_io: return (Offset >= Hexagon_MEMB_OFFSET_MIN) && (Offset <= Hexagon_MEMB_OFFSET_MAX); case Hexagon::A2_addi: return (Offset >= Hexagon_ADDI_OFFSET_MIN) && (Offset <= Hexagon_ADDI_OFFSET_MAX); case Hexagon::L4_iadd_memopw_io: case Hexagon::L4_isub_memopw_io: case Hexagon::L4_add_memopw_io: case Hexagon::L4_sub_memopw_io: case Hexagon::L4_and_memopw_io: case Hexagon::L4_or_memopw_io: return (0 <= Offset && Offset <= 255); case Hexagon::L4_iadd_memoph_io: case Hexagon::L4_isub_memoph_io: case Hexagon::L4_add_memoph_io: case Hexagon::L4_sub_memoph_io: case Hexagon::L4_and_memoph_io: case Hexagon::L4_or_memoph_io: return (0 <= Offset && Offset <= 127); case Hexagon::L4_iadd_memopb_io: case Hexagon::L4_isub_memopb_io: case Hexagon::L4_add_memopb_io: case Hexagon::L4_sub_memopb_io: case Hexagon::L4_and_memopb_io: case Hexagon::L4_or_memopb_io: return (0 <= Offset && Offset <= 63); // LDriw_xxx and STriw_xxx are pseudo operations, so it has to take offset of // any size. Later pass knows how to handle it. case Hexagon::STriw_pred: case Hexagon::LDriw_pred: case Hexagon::STriw_ctr: case Hexagon::LDriw_ctr: return true; case Hexagon::PS_fi: case Hexagon::PS_fia: case Hexagon::INLINEASM: return true; case Hexagon::L2_ploadrbt_io: case Hexagon::L2_ploadrbf_io: case Hexagon::L2_ploadrubt_io: case Hexagon::L2_ploadrubf_io: case Hexagon::S2_pstorerbt_io: case Hexagon::S2_pstorerbf_io: return isUInt<6>(Offset); case Hexagon::L2_ploadrht_io: case Hexagon::L2_ploadrhf_io: case Hexagon::L2_ploadruht_io: case Hexagon::L2_ploadruhf_io: case Hexagon::S2_pstorerht_io: case Hexagon::S2_pstorerhf_io: return isShiftedUInt<6,1>(Offset); case Hexagon::L2_ploadrit_io: case Hexagon::L2_ploadrif_io: case Hexagon::S2_pstorerit_io: case Hexagon::S2_pstorerif_io: return isShiftedUInt<6,2>(Offset); case Hexagon::L2_ploadrdt_io: case Hexagon::L2_ploadrdf_io: case Hexagon::S2_pstorerdt_io: case Hexagon::S2_pstorerdf_io: return isShiftedUInt<6,3>(Offset); } // switch llvm_unreachable("No offset range is defined for this opcode. " "Please define it in the above switch statement!"); } bool HexagonInstrInfo::isVecAcc(const MachineInstr &MI) const { return isHVXVec(MI) && isAccumulator(MI); } bool HexagonInstrInfo::isVecALU(const MachineInstr &MI) const { const uint64_t F = get(MI.getOpcode()).TSFlags; const uint64_t V = ((F >> HexagonII::TypePos) & HexagonII::TypeMask); return V == HexagonII::TypeCVI_VA || V == HexagonII::TypeCVI_VA_DV; } bool HexagonInstrInfo::isVecUsableNextPacket(const MachineInstr &ProdMI, const MachineInstr &ConsMI) const { if (EnableACCForwarding && isVecAcc(ProdMI) && isVecAcc(ConsMI)) return true; if (EnableALUForwarding && (isVecALU(ConsMI) || isLateSourceInstr(ConsMI))) return true; if (mayBeNewStore(ConsMI)) return true; return false; } bool HexagonInstrInfo::isZeroExtendingLoad(const MachineInstr &MI) const { switch (MI.getOpcode()) { // Byte case Hexagon::L2_loadrub_io: case Hexagon::L4_loadrub_ur: case Hexagon::L4_loadrub_ap: case Hexagon::L2_loadrub_pr: case Hexagon::L2_loadrub_pbr: case Hexagon::L2_loadrub_pi: case Hexagon::L2_loadrub_pci: case Hexagon::L2_loadrub_pcr: case Hexagon::L2_loadbzw2_io: case Hexagon::L4_loadbzw2_ur: case Hexagon::L4_loadbzw2_ap: case Hexagon::L2_loadbzw2_pr: case Hexagon::L2_loadbzw2_pbr: case Hexagon::L2_loadbzw2_pi: case Hexagon::L2_loadbzw2_pci: case Hexagon::L2_loadbzw2_pcr: case Hexagon::L2_loadbzw4_io: case Hexagon::L4_loadbzw4_ur: case Hexagon::L4_loadbzw4_ap: case Hexagon::L2_loadbzw4_pr: case Hexagon::L2_loadbzw4_pbr: case Hexagon::L2_loadbzw4_pi: case Hexagon::L2_loadbzw4_pci: case Hexagon::L2_loadbzw4_pcr: case Hexagon::L4_loadrub_rr: case Hexagon::L2_ploadrubt_io: case Hexagon::L2_ploadrubt_pi: case Hexagon::L2_ploadrubf_io: case Hexagon::L2_ploadrubf_pi: case Hexagon::L2_ploadrubtnew_io: case Hexagon::L2_ploadrubfnew_io: case Hexagon::L4_ploadrubt_rr: case Hexagon::L4_ploadrubf_rr: case Hexagon::L4_ploadrubtnew_rr: case Hexagon::L4_ploadrubfnew_rr: case Hexagon::L2_ploadrubtnew_pi: case Hexagon::L2_ploadrubfnew_pi: case Hexagon::L4_ploadrubt_abs: case Hexagon::L4_ploadrubf_abs: case Hexagon::L4_ploadrubtnew_abs: case Hexagon::L4_ploadrubfnew_abs: case Hexagon::L2_loadrubgp: // Half case Hexagon::L2_loadruh_io: case Hexagon::L4_loadruh_ur: case Hexagon::L4_loadruh_ap: case Hexagon::L2_loadruh_pr: case Hexagon::L2_loadruh_pbr: case Hexagon::L2_loadruh_pi: case Hexagon::L2_loadruh_pci: case Hexagon::L2_loadruh_pcr: case Hexagon::L4_loadruh_rr: case Hexagon::L2_ploadruht_io: case Hexagon::L2_ploadruht_pi: case Hexagon::L2_ploadruhf_io: case Hexagon::L2_ploadruhf_pi: case Hexagon::L2_ploadruhtnew_io: case Hexagon::L2_ploadruhfnew_io: case Hexagon::L4_ploadruht_rr: case Hexagon::L4_ploadruhf_rr: case Hexagon::L4_ploadruhtnew_rr: case Hexagon::L4_ploadruhfnew_rr: case Hexagon::L2_ploadruhtnew_pi: case Hexagon::L2_ploadruhfnew_pi: case Hexagon::L4_ploadruht_abs: case Hexagon::L4_ploadruhf_abs: case Hexagon::L4_ploadruhtnew_abs: case Hexagon::L4_ploadruhfnew_abs: case Hexagon::L2_loadruhgp: return true; default: return false; } } // Add latency to instruction. bool HexagonInstrInfo::addLatencyToSchedule(const MachineInstr &MI1, const MachineInstr &MI2) const { if (isHVXVec(MI1) && isHVXVec(MI2)) if (!isVecUsableNextPacket(MI1, MI2)) return true; return false; } /// Get the base register and byte offset of a load/store instr. bool HexagonInstrInfo::getMemOperandsWithOffsetWidth( const MachineInstr &LdSt, SmallVectorImpl &BaseOps, int64_t &Offset, bool &OffsetIsScalable, unsigned &Width, const TargetRegisterInfo *TRI) const { OffsetIsScalable = false; const MachineOperand *BaseOp = getBaseAndOffset(LdSt, Offset, Width); if (!BaseOp || !BaseOp->isReg()) return false; BaseOps.push_back(BaseOp); return true; } /// Can these instructions execute at the same time in a bundle. bool HexagonInstrInfo::canExecuteInBundle(const MachineInstr &First, const MachineInstr &Second) const { if (Second.mayStore() && First.getOpcode() == Hexagon::S2_allocframe) { const MachineOperand &Op = Second.getOperand(0); if (Op.isReg() && Op.isUse() && Op.getReg() == Hexagon::R29) return true; } if (DisableNVSchedule) return false; if (mayBeNewStore(Second)) { // Make sure the definition of the first instruction is the value being // stored. const MachineOperand &Stored = Second.getOperand(Second.getNumOperands() - 1); if (!Stored.isReg()) return false; for (unsigned i = 0, e = First.getNumOperands(); i < e; ++i) { const MachineOperand &Op = First.getOperand(i); if (Op.isReg() && Op.isDef() && Op.getReg() == Stored.getReg()) return true; } } return false; } bool HexagonInstrInfo::doesNotReturn(const MachineInstr &CallMI) const { unsigned Opc = CallMI.getOpcode(); return Opc == Hexagon::PS_call_nr || Opc == Hexagon::PS_callr_nr; } bool HexagonInstrInfo::hasEHLabel(const MachineBasicBlock *B) const { for (auto &I : *B) if (I.isEHLabel()) return true; return false; } // Returns true if an instruction can be converted into a non-extended // equivalent instruction. bool HexagonInstrInfo::hasNonExtEquivalent(const MachineInstr &MI) const { short NonExtOpcode; // Check if the instruction has a register form that uses register in place // of the extended operand, if so return that as the non-extended form. if (Hexagon::getRegForm(MI.getOpcode()) >= 0) return true; if (MI.getDesc().mayLoad() || MI.getDesc().mayStore()) { // Check addressing mode and retrieve non-ext equivalent instruction. switch (getAddrMode(MI)) { case HexagonII::Absolute: // Load/store with absolute addressing mode can be converted into // base+offset mode. NonExtOpcode = Hexagon::changeAddrMode_abs_io(MI.getOpcode()); break; case HexagonII::BaseImmOffset: // Load/store with base+offset addressing mode can be converted into // base+register offset addressing mode. However left shift operand should // be set to 0. NonExtOpcode = Hexagon::changeAddrMode_io_rr(MI.getOpcode()); break; case HexagonII::BaseLongOffset: NonExtOpcode = Hexagon::changeAddrMode_ur_rr(MI.getOpcode()); break; default: return false; } if (NonExtOpcode < 0) return false; return true; } return false; } bool HexagonInstrInfo::hasPseudoInstrPair(const MachineInstr &MI) const { return Hexagon::getRealHWInstr(MI.getOpcode(), Hexagon::InstrType_Pseudo) >= 0; } bool HexagonInstrInfo::hasUncondBranch(const MachineBasicBlock *B) const { MachineBasicBlock::const_iterator I = B->getFirstTerminator(), E = B->end(); while (I != E) { if (I->isBarrier()) return true; ++I; } return false; } // Returns true, if a LD insn can be promoted to a cur load. bool HexagonInstrInfo::mayBeCurLoad(const MachineInstr &MI) const { const uint64_t F = MI.getDesc().TSFlags; return ((F >> HexagonII::mayCVLoadPos) & HexagonII::mayCVLoadMask) && Subtarget.hasV60Ops(); } // Returns true, if a ST insn can be promoted to a new-value store. bool HexagonInstrInfo::mayBeNewStore(const MachineInstr &MI) const { if (MI.mayStore() && !Subtarget.useNewValueStores()) return false; const uint64_t F = MI.getDesc().TSFlags; return (F >> HexagonII::mayNVStorePos) & HexagonII::mayNVStoreMask; } bool HexagonInstrInfo::producesStall(const MachineInstr &ProdMI, const MachineInstr &ConsMI) const { // There is no stall when ProdMI is not a V60 vector. if (!isHVXVec(ProdMI)) return false; // There is no stall when ProdMI and ConsMI are not dependent. if (!isDependent(ProdMI, ConsMI)) return false; // When Forward Scheduling is enabled, there is no stall if ProdMI and ConsMI // are scheduled in consecutive packets. if (isVecUsableNextPacket(ProdMI, ConsMI)) return false; return true; } bool HexagonInstrInfo::producesStall(const MachineInstr &MI, MachineBasicBlock::const_instr_iterator BII) const { // There is no stall when I is not a V60 vector. if (!isHVXVec(MI)) return false; MachineBasicBlock::const_instr_iterator MII = BII; MachineBasicBlock::const_instr_iterator MIE = MII->getParent()->instr_end(); if (!MII->isBundle()) return producesStall(*MII, MI); for (++MII; MII != MIE && MII->isInsideBundle(); ++MII) { const MachineInstr &J = *MII; if (producesStall(J, MI)) return true; } return false; } bool HexagonInstrInfo::predCanBeUsedAsDotNew(const MachineInstr &MI, unsigned PredReg) const { for (const MachineOperand &MO : MI.operands()) { // Predicate register must be explicitly defined. if (MO.isRegMask() && MO.clobbersPhysReg(PredReg)) return false; if (MO.isReg() && MO.isDef() && MO.isImplicit() && (MO.getReg() == PredReg)) return false; } // Instruction that produce late predicate cannot be used as sources of // dot-new. switch (MI.getOpcode()) { case Hexagon::A4_addp_c: case Hexagon::A4_subp_c: case Hexagon::A4_tlbmatch: case Hexagon::A5_ACS: case Hexagon::F2_sfinvsqrta: case Hexagon::F2_sfrecipa: case Hexagon::J2_endloop0: case Hexagon::J2_endloop01: case Hexagon::J2_ploop1si: case Hexagon::J2_ploop1sr: case Hexagon::J2_ploop2si: case Hexagon::J2_ploop2sr: case Hexagon::J2_ploop3si: case Hexagon::J2_ploop3sr: case Hexagon::S2_cabacdecbin: case Hexagon::S2_storew_locked: case Hexagon::S4_stored_locked: return false; } return true; } bool HexagonInstrInfo::PredOpcodeHasJMP_c(unsigned Opcode) const { return Opcode == Hexagon::J2_jumpt || Opcode == Hexagon::J2_jumptpt || Opcode == Hexagon::J2_jumpf || Opcode == Hexagon::J2_jumpfpt || Opcode == Hexagon::J2_jumptnew || Opcode == Hexagon::J2_jumpfnew || Opcode == Hexagon::J2_jumptnewpt || Opcode == Hexagon::J2_jumpfnewpt; } bool HexagonInstrInfo::predOpcodeHasNot(ArrayRef Cond) const { if (Cond.empty() || !isPredicated(Cond[0].getImm())) return false; return !isPredicatedTrue(Cond[0].getImm()); } unsigned HexagonInstrInfo::getAddrMode(const MachineInstr &MI) const { const uint64_t F = MI.getDesc().TSFlags; return (F >> HexagonII::AddrModePos) & HexagonII::AddrModeMask; } // Returns the base register in a memory access (load/store). The offset is // returned in Offset and the access size is returned in AccessSize. // If the base operand has a subregister or the offset field does not contain // an immediate value, return nullptr. MachineOperand *HexagonInstrInfo::getBaseAndOffset(const MachineInstr &MI, int64_t &Offset, unsigned &AccessSize) const { // Return if it is not a base+offset type instruction or a MemOp. if (getAddrMode(MI) != HexagonII::BaseImmOffset && getAddrMode(MI) != HexagonII::BaseLongOffset && !isMemOp(MI) && !isPostIncrement(MI)) return nullptr; AccessSize = getMemAccessSize(MI); unsigned BasePos = 0, OffsetPos = 0; if (!getBaseAndOffsetPosition(MI, BasePos, OffsetPos)) return nullptr; // Post increment updates its EA after the mem access, // so we need to treat its offset as zero. if (isPostIncrement(MI)) { Offset = 0; } else { const MachineOperand &OffsetOp = MI.getOperand(OffsetPos); if (!OffsetOp.isImm()) return nullptr; Offset = OffsetOp.getImm(); } const MachineOperand &BaseOp = MI.getOperand(BasePos); if (BaseOp.getSubReg() != 0) return nullptr; return &const_cast(BaseOp); } /// Return the position of the base and offset operands for this instruction. bool HexagonInstrInfo::getBaseAndOffsetPosition(const MachineInstr &MI, unsigned &BasePos, unsigned &OffsetPos) const { if (!isAddrModeWithOffset(MI) && !isPostIncrement(MI)) return false; // Deal with memops first. if (isMemOp(MI)) { BasePos = 0; OffsetPos = 1; } else if (MI.mayStore()) { BasePos = 0; OffsetPos = 1; } else if (MI.mayLoad()) { BasePos = 1; OffsetPos = 2; } else return false; if (isPredicated(MI)) { BasePos++; OffsetPos++; } if (isPostIncrement(MI)) { BasePos++; OffsetPos++; } if (!MI.getOperand(BasePos).isReg() || !MI.getOperand(OffsetPos).isImm()) return false; return true; } // Inserts branching instructions in reverse order of their occurrence. // e.g. jump_t t1 (i1) // jump t2 (i2) // Jumpers = {i2, i1} SmallVector HexagonInstrInfo::getBranchingInstrs( MachineBasicBlock& MBB) const { SmallVector Jumpers; // If the block has no terminators, it just falls into the block after it. MachineBasicBlock::instr_iterator I = MBB.instr_end(); if (I == MBB.instr_begin()) return Jumpers; // A basic block may looks like this: // // [ insn // EH_LABEL // insn // insn // insn // EH_LABEL // insn ] // // It has two succs but does not have a terminator // Don't know how to handle it. do { --I; if (I->isEHLabel()) return Jumpers; } while (I != MBB.instr_begin()); I = MBB.instr_end(); --I; while (I->isDebugInstr()) { if (I == MBB.instr_begin()) return Jumpers; --I; } if (!isUnpredicatedTerminator(*I)) return Jumpers; // Get the last instruction in the block. MachineInstr *LastInst = &*I; Jumpers.push_back(LastInst); MachineInstr *SecondLastInst = nullptr; // Find one more terminator if present. do { if (&*I != LastInst && !I->isBundle() && isUnpredicatedTerminator(*I)) { if (!SecondLastInst) { SecondLastInst = &*I; Jumpers.push_back(SecondLastInst); } else // This is a third branch. return Jumpers; } if (I == MBB.instr_begin()) break; --I; } while (true); return Jumpers; } // Returns Operand Index for the constant extended instruction. unsigned HexagonInstrInfo::getCExtOpNum(const MachineInstr &MI) const { const uint64_t F = MI.getDesc().TSFlags; return (F >> HexagonII::ExtendableOpPos) & HexagonII::ExtendableOpMask; } // See if instruction could potentially be a duplex candidate. // If so, return its group. Zero otherwise. HexagonII::CompoundGroup HexagonInstrInfo::getCompoundCandidateGroup( const MachineInstr &MI) const { unsigned DstReg, SrcReg, Src1Reg, Src2Reg; switch (MI.getOpcode()) { default: return HexagonII::HCG_None; // // Compound pairs. // "p0=cmp.eq(Rs16,Rt16); if (p0.new) jump:nt #r9:2" // "Rd16=#U6 ; jump #r9:2" // "Rd16=Rs16 ; jump #r9:2" // case Hexagon::C2_cmpeq: case Hexagon::C2_cmpgt: case Hexagon::C2_cmpgtu: DstReg = MI.getOperand(0).getReg(); Src1Reg = MI.getOperand(1).getReg(); Src2Reg = MI.getOperand(2).getReg(); if (Hexagon::PredRegsRegClass.contains(DstReg) && (Hexagon::P0 == DstReg || Hexagon::P1 == DstReg) && isIntRegForSubInst(Src1Reg) && isIntRegForSubInst(Src2Reg)) return HexagonII::HCG_A; break; case Hexagon::C2_cmpeqi: case Hexagon::C2_cmpgti: case Hexagon::C2_cmpgtui: // P0 = cmp.eq(Rs,#u2) DstReg = MI.getOperand(0).getReg(); SrcReg = MI.getOperand(1).getReg(); if (Hexagon::PredRegsRegClass.contains(DstReg) && (Hexagon::P0 == DstReg || Hexagon::P1 == DstReg) && isIntRegForSubInst(SrcReg) && MI.getOperand(2).isImm() && ((isUInt<5>(MI.getOperand(2).getImm())) || (MI.getOperand(2).getImm() == -1))) return HexagonII::HCG_A; break; case Hexagon::A2_tfr: // Rd = Rs DstReg = MI.getOperand(0).getReg(); SrcReg = MI.getOperand(1).getReg(); if (isIntRegForSubInst(DstReg) && isIntRegForSubInst(SrcReg)) return HexagonII::HCG_A; break; case Hexagon::A2_tfrsi: // Rd = #u6 // Do not test for #u6 size since the const is getting extended // regardless and compound could be formed. DstReg = MI.getOperand(0).getReg(); if (isIntRegForSubInst(DstReg)) return HexagonII::HCG_A; break; case Hexagon::S2_tstbit_i: DstReg = MI.getOperand(0).getReg(); Src1Reg = MI.getOperand(1).getReg(); if (Hexagon::PredRegsRegClass.contains(DstReg) && (Hexagon::P0 == DstReg || Hexagon::P1 == DstReg) && MI.getOperand(2).isImm() && isIntRegForSubInst(Src1Reg) && (MI.getOperand(2).getImm() == 0)) return HexagonII::HCG_A; break; // The fact that .new form is used pretty much guarantees // that predicate register will match. Nevertheless, // there could be some false positives without additional // checking. case Hexagon::J2_jumptnew: case Hexagon::J2_jumpfnew: case Hexagon::J2_jumptnewpt: case Hexagon::J2_jumpfnewpt: Src1Reg = MI.getOperand(0).getReg(); if (Hexagon::PredRegsRegClass.contains(Src1Reg) && (Hexagon::P0 == Src1Reg || Hexagon::P1 == Src1Reg)) return HexagonII::HCG_B; break; // Transfer and jump: // Rd=#U6 ; jump #r9:2 // Rd=Rs ; jump #r9:2 // Do not test for jump range here. case Hexagon::J2_jump: case Hexagon::RESTORE_DEALLOC_RET_JMP_V4: case Hexagon::RESTORE_DEALLOC_RET_JMP_V4_PIC: return HexagonII::HCG_C; } return HexagonII::HCG_None; } // Returns -1 when there is no opcode found. unsigned HexagonInstrInfo::getCompoundOpcode(const MachineInstr &GA, const MachineInstr &GB) const { assert(getCompoundCandidateGroup(GA) == HexagonII::HCG_A); assert(getCompoundCandidateGroup(GB) == HexagonII::HCG_B); if ((GA.getOpcode() != Hexagon::C2_cmpeqi) || (GB.getOpcode() != Hexagon::J2_jumptnew)) return -1u; Register DestReg = GA.getOperand(0).getReg(); if (!GB.readsRegister(DestReg)) return -1u; if (DestReg != Hexagon::P0 && DestReg != Hexagon::P1) return -1u; // The value compared against must be either u5 or -1. const MachineOperand &CmpOp = GA.getOperand(2); if (!CmpOp.isImm()) return -1u; int V = CmpOp.getImm(); if (V == -1) return DestReg == Hexagon::P0 ? Hexagon::J4_cmpeqn1_tp0_jump_nt : Hexagon::J4_cmpeqn1_tp1_jump_nt; if (!isUInt<5>(V)) return -1u; return DestReg == Hexagon::P0 ? Hexagon::J4_cmpeqi_tp0_jump_nt : Hexagon::J4_cmpeqi_tp1_jump_nt; } // Returns -1 if there is no opcode found. int HexagonInstrInfo::getDuplexOpcode(const MachineInstr &MI, bool ForBigCore) const { // Static table to switch the opcodes across Tiny Core and Big Core. // dup_ opcodes are Big core opcodes. // NOTE: There are special instructions that need to handled later. // L4_return* instructions, they will only occupy SLOT0 (on big core too). // PS_jmpret - This pseudo translates to J2_jumpr which occupies only SLOT2. // The compiler need to base the root instruction to L6_return_map_to_raw // which can go any slot. static const std::map DupMap = { {Hexagon::A2_add, Hexagon::dup_A2_add}, {Hexagon::A2_addi, Hexagon::dup_A2_addi}, {Hexagon::A2_andir, Hexagon::dup_A2_andir}, {Hexagon::A2_combineii, Hexagon::dup_A2_combineii}, {Hexagon::A2_sxtb, Hexagon::dup_A2_sxtb}, {Hexagon::A2_sxth, Hexagon::dup_A2_sxth}, {Hexagon::A2_tfr, Hexagon::dup_A2_tfr}, {Hexagon::A2_tfrsi, Hexagon::dup_A2_tfrsi}, {Hexagon::A2_zxtb, Hexagon::dup_A2_zxtb}, {Hexagon::A2_zxth, Hexagon::dup_A2_zxth}, {Hexagon::A4_combineii, Hexagon::dup_A4_combineii}, {Hexagon::A4_combineir, Hexagon::dup_A4_combineir}, {Hexagon::A4_combineri, Hexagon::dup_A4_combineri}, {Hexagon::C2_cmoveif, Hexagon::dup_C2_cmoveif}, {Hexagon::C2_cmoveit, Hexagon::dup_C2_cmoveit}, {Hexagon::C2_cmovenewif, Hexagon::dup_C2_cmovenewif}, {Hexagon::C2_cmovenewit, Hexagon::dup_C2_cmovenewit}, {Hexagon::C2_cmpeqi, Hexagon::dup_C2_cmpeqi}, {Hexagon::L2_deallocframe, Hexagon::dup_L2_deallocframe}, {Hexagon::L2_loadrb_io, Hexagon::dup_L2_loadrb_io}, {Hexagon::L2_loadrd_io, Hexagon::dup_L2_loadrd_io}, {Hexagon::L2_loadrh_io, Hexagon::dup_L2_loadrh_io}, {Hexagon::L2_loadri_io, Hexagon::dup_L2_loadri_io}, {Hexagon::L2_loadrub_io, Hexagon::dup_L2_loadrub_io}, {Hexagon::L2_loadruh_io, Hexagon::dup_L2_loadruh_io}, {Hexagon::S2_allocframe, Hexagon::dup_S2_allocframe}, {Hexagon::S2_storerb_io, Hexagon::dup_S2_storerb_io}, {Hexagon::S2_storerd_io, Hexagon::dup_S2_storerd_io}, {Hexagon::S2_storerh_io, Hexagon::dup_S2_storerh_io}, {Hexagon::S2_storeri_io, Hexagon::dup_S2_storeri_io}, {Hexagon::S4_storeirb_io, Hexagon::dup_S4_storeirb_io}, {Hexagon::S4_storeiri_io, Hexagon::dup_S4_storeiri_io}, }; unsigned OpNum = MI.getOpcode(); // Conversion to Big core. if (ForBigCore) { auto Iter = DupMap.find(OpNum); if (Iter != DupMap.end()) return Iter->second; } else { // Conversion to Tiny core. for (auto Iter = DupMap.begin(), End = DupMap.end(); Iter != End; ++Iter) if (Iter->second == OpNum) return Iter->first; } return -1; } int HexagonInstrInfo::getCondOpcode(int Opc, bool invertPredicate) const { enum Hexagon::PredSense inPredSense; inPredSense = invertPredicate ? Hexagon::PredSense_false : Hexagon::PredSense_true; int CondOpcode = Hexagon::getPredOpcode(Opc, inPredSense); if (CondOpcode >= 0) // Valid Conditional opcode/instruction return CondOpcode; llvm_unreachable("Unexpected predicable instruction"); } // Return the cur value instruction for a given store. int HexagonInstrInfo::getDotCurOp(const MachineInstr &MI) const { switch (MI.getOpcode()) { default: llvm_unreachable("Unknown .cur type"); case Hexagon::V6_vL32b_pi: return Hexagon::V6_vL32b_cur_pi; case Hexagon::V6_vL32b_ai: return Hexagon::V6_vL32b_cur_ai; case Hexagon::V6_vL32b_nt_pi: return Hexagon::V6_vL32b_nt_cur_pi; case Hexagon::V6_vL32b_nt_ai: return Hexagon::V6_vL32b_nt_cur_ai; } return 0; } // Return the regular version of the .cur instruction. int HexagonInstrInfo::getNonDotCurOp(const MachineInstr &MI) const { switch (MI.getOpcode()) { default: llvm_unreachable("Unknown .cur type"); case Hexagon::V6_vL32b_cur_pi: return Hexagon::V6_vL32b_pi; case Hexagon::V6_vL32b_cur_ai: return Hexagon::V6_vL32b_ai; case Hexagon::V6_vL32b_nt_cur_pi: return Hexagon::V6_vL32b_nt_pi; case Hexagon::V6_vL32b_nt_cur_ai: return Hexagon::V6_vL32b_nt_ai; } return 0; } // The diagram below shows the steps involved in the conversion of a predicated // store instruction to its .new predicated new-value form. // // Note: It doesn't include conditional new-value stores as they can't be // converted to .new predicate. // // p.new NV store [ if(p0.new)memw(R0+#0)=R2.new ] // ^ ^ // / \ (not OK. it will cause new-value store to be // / X conditional on p0.new while R2 producer is // / \ on p0) // / \. // p.new store p.old NV store // [if(p0.new)memw(R0+#0)=R2] [if(p0)memw(R0+#0)=R2.new] // ^ ^ // \ / // \ / // \ / // p.old store // [if (p0)memw(R0+#0)=R2] // // The following set of instructions further explains the scenario where // conditional new-value store becomes invalid when promoted to .new predicate // form. // // { 1) if (p0) r0 = add(r1, r2) // 2) p0 = cmp.eq(r3, #0) } // // 3) if (p0) memb(r1+#0) = r0 --> this instruction can't be grouped with // the first two instructions because in instr 1, r0 is conditional on old value // of p0 but its use in instr 3 is conditional on p0 modified by instr 2 which // is not valid for new-value stores. // Predicated new value stores (i.e. if (p0) memw(..)=r0.new) are excluded // from the "Conditional Store" list. Because a predicated new value store // would NOT be promoted to a double dot new store. See diagram below: // This function returns yes for those stores that are predicated but not // yet promoted to predicate dot new instructions. // // +---------------------+ // /-----| if (p0) memw(..)=r0 |---------\~ // || +---------------------+ || // promote || /\ /\ || promote // || /||\ /||\ || // \||/ demote || \||/ // \/ || || \/ // +-------------------------+ || +-------------------------+ // | if (p0.new) memw(..)=r0 | || | if (p0) memw(..)=r0.new | // +-------------------------+ || +-------------------------+ // || || || // || demote \||/ // promote || \/ NOT possible // || || /\~ // \||/ || /||\~ // \/ || || // +-----------------------------+ // | if (p0.new) memw(..)=r0.new | // +-----------------------------+ // Double Dot New Store // // Returns the most basic instruction for the .new predicated instructions and // new-value stores. // For example, all of the following instructions will be converted back to the // same instruction: // 1) if (p0.new) memw(R0+#0) = R1.new ---> // 2) if (p0) memw(R0+#0)= R1.new -------> if (p0) memw(R0+#0) = R1 // 3) if (p0.new) memw(R0+#0) = R1 ---> // // To understand the translation of instruction 1 to its original form, consider // a packet with 3 instructions. // { p0 = cmp.eq(R0,R1) // if (p0.new) R2 = add(R3, R4) // R5 = add (R3, R1) // } // if (p0) memw(R5+#0) = R2 <--- trying to include it in the previous packet // // This instruction can be part of the previous packet only if both p0 and R2 // are promoted to .new values. This promotion happens in steps, first // predicate register is promoted to .new and in the next iteration R2 is // promoted. Therefore, in case of dependence check failure (due to R5) during // next iteration, it should be converted back to its most basic form. // Return the new value instruction for a given store. int HexagonInstrInfo::getDotNewOp(const MachineInstr &MI) const { int NVOpcode = Hexagon::getNewValueOpcode(MI.getOpcode()); if (NVOpcode >= 0) // Valid new-value store instruction. return NVOpcode; switch (MI.getOpcode()) { default: report_fatal_error(std::string("Unknown .new type: ") + std::to_string(MI.getOpcode())); case Hexagon::S4_storerb_ur: return Hexagon::S4_storerbnew_ur; case Hexagon::S2_storerb_pci: return Hexagon::S2_storerb_pci; case Hexagon::S2_storeri_pci: return Hexagon::S2_storeri_pci; case Hexagon::S2_storerh_pci: return Hexagon::S2_storerh_pci; case Hexagon::S2_storerd_pci: return Hexagon::S2_storerd_pci; case Hexagon::S2_storerf_pci: return Hexagon::S2_storerf_pci; case Hexagon::V6_vS32b_ai: return Hexagon::V6_vS32b_new_ai; case Hexagon::V6_vS32b_pi: return Hexagon::V6_vS32b_new_pi; } return 0; } // Returns the opcode to use when converting MI, which is a conditional jump, // into a conditional instruction which uses the .new value of the predicate. // We also use branch probabilities to add a hint to the jump. // If MBPI is null, all edges will be treated as equally likely for the // purposes of establishing a predication hint. int HexagonInstrInfo::getDotNewPredJumpOp(const MachineInstr &MI, const MachineBranchProbabilityInfo *MBPI) const { // We assume that block can have at most two successors. const MachineBasicBlock *Src = MI.getParent(); const MachineOperand &BrTarget = MI.getOperand(1); bool Taken = false; const BranchProbability OneHalf(1, 2); auto getEdgeProbability = [MBPI] (const MachineBasicBlock *Src, const MachineBasicBlock *Dst) { if (MBPI) return MBPI->getEdgeProbability(Src, Dst); return BranchProbability(1, Src->succ_size()); }; if (BrTarget.isMBB()) { const MachineBasicBlock *Dst = BrTarget.getMBB(); Taken = getEdgeProbability(Src, Dst) >= OneHalf; } else { // The branch target is not a basic block (most likely a function). // Since BPI only gives probabilities for targets that are basic blocks, // try to identify another target of this branch (potentially a fall- // -through) and check the probability of that target. // // The only handled branch combinations are: // - one conditional branch, // - one conditional branch followed by one unconditional branch. // Otherwise, assume not-taken. assert(MI.isConditionalBranch()); const MachineBasicBlock &B = *MI.getParent(); bool SawCond = false, Bad = false; for (const MachineInstr &I : B) { if (!I.isBranch()) continue; if (I.isConditionalBranch()) { SawCond = true; if (&I != &MI) { Bad = true; break; } } if (I.isUnconditionalBranch() && !SawCond) { Bad = true; break; } } if (!Bad) { MachineBasicBlock::const_instr_iterator It(MI); MachineBasicBlock::const_instr_iterator NextIt = std::next(It); if (NextIt == B.instr_end()) { // If this branch is the last, look for the fall-through block. for (const MachineBasicBlock *SB : B.successors()) { if (!B.isLayoutSuccessor(SB)) continue; Taken = getEdgeProbability(Src, SB) < OneHalf; break; } } else { assert(NextIt->isUnconditionalBranch()); // Find the first MBB operand and assume it's the target. const MachineBasicBlock *BT = nullptr; for (const MachineOperand &Op : NextIt->operands()) { if (!Op.isMBB()) continue; BT = Op.getMBB(); break; } Taken = BT && getEdgeProbability(Src, BT) < OneHalf; } } // if (!Bad) } // The Taken flag should be set to something reasonable by this point. switch (MI.getOpcode()) { case Hexagon::J2_jumpt: return Taken ? Hexagon::J2_jumptnewpt : Hexagon::J2_jumptnew; case Hexagon::J2_jumpf: return Taken ? Hexagon::J2_jumpfnewpt : Hexagon::J2_jumpfnew; default: llvm_unreachable("Unexpected jump instruction."); } } // Return .new predicate version for an instruction. int HexagonInstrInfo::getDotNewPredOp(const MachineInstr &MI, const MachineBranchProbabilityInfo *MBPI) const { switch (MI.getOpcode()) { // Condtional Jumps case Hexagon::J2_jumpt: case Hexagon::J2_jumpf: return getDotNewPredJumpOp(MI, MBPI); } int NewOpcode = Hexagon::getPredNewOpcode(MI.getOpcode()); if (NewOpcode >= 0) return NewOpcode; return 0; } int HexagonInstrInfo::getDotOldOp(const MachineInstr &MI) const { int NewOp = MI.getOpcode(); if (isPredicated(NewOp) && isPredicatedNew(NewOp)) { // Get predicate old form NewOp = Hexagon::getPredOldOpcode(NewOp); // All Hexagon architectures have prediction bits on dot-new branches, // but only Hexagon V60+ has prediction bits on dot-old ones. Make sure // to pick the right opcode when converting back to dot-old. if (!Subtarget.getFeatureBits()[Hexagon::ArchV60]) { switch (NewOp) { case Hexagon::J2_jumptpt: NewOp = Hexagon::J2_jumpt; break; case Hexagon::J2_jumpfpt: NewOp = Hexagon::J2_jumpf; break; case Hexagon::J2_jumprtpt: NewOp = Hexagon::J2_jumprt; break; case Hexagon::J2_jumprfpt: NewOp = Hexagon::J2_jumprf; break; } } assert(NewOp >= 0 && "Couldn't change predicate new instruction to its old form."); } if (isNewValueStore(NewOp)) { // Convert into non-new-value format NewOp = Hexagon::getNonNVStore(NewOp); assert(NewOp >= 0 && "Couldn't change new-value store to its old form."); } if (Subtarget.hasV60Ops()) return NewOp; // Subtargets prior to V60 didn't support 'taken' forms of predicated jumps. switch (NewOp) { case Hexagon::J2_jumpfpt: return Hexagon::J2_jumpf; case Hexagon::J2_jumptpt: return Hexagon::J2_jumpt; case Hexagon::J2_jumprfpt: return Hexagon::J2_jumprf; case Hexagon::J2_jumprtpt: return Hexagon::J2_jumprt; } return NewOp; } // See if instruction could potentially be a duplex candidate. // If so, return its group. Zero otherwise. HexagonII::SubInstructionGroup HexagonInstrInfo::getDuplexCandidateGroup( const MachineInstr &MI) const { unsigned DstReg, SrcReg, Src1Reg, Src2Reg; const HexagonRegisterInfo &HRI = *Subtarget.getRegisterInfo(); switch (MI.getOpcode()) { default: return HexagonII::HSIG_None; // // Group L1: // // Rd = memw(Rs+#u4:2) // Rd = memub(Rs+#u4:0) case Hexagon::L2_loadri_io: case Hexagon::dup_L2_loadri_io: DstReg = MI.getOperand(0).getReg(); SrcReg = MI.getOperand(1).getReg(); // Special case this one from Group L2. // Rd = memw(r29+#u5:2) if (isIntRegForSubInst(DstReg)) { if (Hexagon::IntRegsRegClass.contains(SrcReg) && HRI.getStackRegister() == SrcReg && MI.getOperand(2).isImm() && isShiftedUInt<5,2>(MI.getOperand(2).getImm())) return HexagonII::HSIG_L2; // Rd = memw(Rs+#u4:2) if (isIntRegForSubInst(SrcReg) && (MI.getOperand(2).isImm() && isShiftedUInt<4,2>(MI.getOperand(2).getImm()))) return HexagonII::HSIG_L1; } break; case Hexagon::L2_loadrub_io: case Hexagon::dup_L2_loadrub_io: // Rd = memub(Rs+#u4:0) DstReg = MI.getOperand(0).getReg(); SrcReg = MI.getOperand(1).getReg(); if (isIntRegForSubInst(DstReg) && isIntRegForSubInst(SrcReg) && MI.getOperand(2).isImm() && isUInt<4>(MI.getOperand(2).getImm())) return HexagonII::HSIG_L1; break; // // Group L2: // // Rd = memh/memuh(Rs+#u3:1) // Rd = memb(Rs+#u3:0) // Rd = memw(r29+#u5:2) - Handled above. // Rdd = memd(r29+#u5:3) // deallocframe // [if ([!]p0[.new])] dealloc_return // [if ([!]p0[.new])] jumpr r31 case Hexagon::L2_loadrh_io: case Hexagon::L2_loadruh_io: case Hexagon::dup_L2_loadrh_io: case Hexagon::dup_L2_loadruh_io: // Rd = memh/memuh(Rs+#u3:1) DstReg = MI.getOperand(0).getReg(); SrcReg = MI.getOperand(1).getReg(); if (isIntRegForSubInst(DstReg) && isIntRegForSubInst(SrcReg) && MI.getOperand(2).isImm() && isShiftedUInt<3,1>(MI.getOperand(2).getImm())) return HexagonII::HSIG_L2; break; case Hexagon::L2_loadrb_io: case Hexagon::dup_L2_loadrb_io: // Rd = memb(Rs+#u3:0) DstReg = MI.getOperand(0).getReg(); SrcReg = MI.getOperand(1).getReg(); if (isIntRegForSubInst(DstReg) && isIntRegForSubInst(SrcReg) && MI.getOperand(2).isImm() && isUInt<3>(MI.getOperand(2).getImm())) return HexagonII::HSIG_L2; break; case Hexagon::L2_loadrd_io: case Hexagon::dup_L2_loadrd_io: // Rdd = memd(r29+#u5:3) DstReg = MI.getOperand(0).getReg(); SrcReg = MI.getOperand(1).getReg(); if (isDblRegForSubInst(DstReg, HRI) && Hexagon::IntRegsRegClass.contains(SrcReg) && HRI.getStackRegister() == SrcReg && MI.getOperand(2).isImm() && isShiftedUInt<5,3>(MI.getOperand(2).getImm())) return HexagonII::HSIG_L2; break; // dealloc_return is not documented in Hexagon Manual, but marked // with A_SUBINSN attribute in iset_v4classic.py. case Hexagon::RESTORE_DEALLOC_RET_JMP_V4: case Hexagon::RESTORE_DEALLOC_RET_JMP_V4_PIC: case Hexagon::L4_return: case Hexagon::L2_deallocframe: case Hexagon::dup_L2_deallocframe: return HexagonII::HSIG_L2; case Hexagon::EH_RETURN_JMPR: case Hexagon::PS_jmpret: case Hexagon::SL2_jumpr31: // jumpr r31 // Actual form JMPR implicit-def %pc, implicit %r31, implicit internal %r0 DstReg = MI.getOperand(0).getReg(); if (Hexagon::IntRegsRegClass.contains(DstReg) && (Hexagon::R31 == DstReg)) return HexagonII::HSIG_L2; break; case Hexagon::PS_jmprett: case Hexagon::PS_jmpretf: case Hexagon::PS_jmprettnewpt: case Hexagon::PS_jmpretfnewpt: case Hexagon::PS_jmprettnew: case Hexagon::PS_jmpretfnew: case Hexagon::SL2_jumpr31_t: case Hexagon::SL2_jumpr31_f: case Hexagon::SL2_jumpr31_tnew: case Hexagon::SL2_jumpr31_fnew: DstReg = MI.getOperand(1).getReg(); SrcReg = MI.getOperand(0).getReg(); // [if ([!]p0[.new])] jumpr r31 if ((Hexagon::PredRegsRegClass.contains(SrcReg) && (Hexagon::P0 == SrcReg)) && (Hexagon::IntRegsRegClass.contains(DstReg) && (Hexagon::R31 == DstReg))) return HexagonII::HSIG_L2; break; case Hexagon::L4_return_t: case Hexagon::L4_return_f: case Hexagon::L4_return_tnew_pnt: case Hexagon::L4_return_fnew_pnt: case Hexagon::L4_return_tnew_pt: case Hexagon::L4_return_fnew_pt: // [if ([!]p0[.new])] dealloc_return SrcReg = MI.getOperand(0).getReg(); if (Hexagon::PredRegsRegClass.contains(SrcReg) && (Hexagon::P0 == SrcReg)) return HexagonII::HSIG_L2; break; // // Group S1: // // memw(Rs+#u4:2) = Rt // memb(Rs+#u4:0) = Rt case Hexagon::S2_storeri_io: case Hexagon::dup_S2_storeri_io: // Special case this one from Group S2. // memw(r29+#u5:2) = Rt Src1Reg = MI.getOperand(0).getReg(); Src2Reg = MI.getOperand(2).getReg(); if (Hexagon::IntRegsRegClass.contains(Src1Reg) && isIntRegForSubInst(Src2Reg) && HRI.getStackRegister() == Src1Reg && MI.getOperand(1).isImm() && isShiftedUInt<5,2>(MI.getOperand(1).getImm())) return HexagonII::HSIG_S2; // memw(Rs+#u4:2) = Rt if (isIntRegForSubInst(Src1Reg) && isIntRegForSubInst(Src2Reg) && MI.getOperand(1).isImm() && isShiftedUInt<4,2>(MI.getOperand(1).getImm())) return HexagonII::HSIG_S1; break; case Hexagon::S2_storerb_io: case Hexagon::dup_S2_storerb_io: // memb(Rs+#u4:0) = Rt Src1Reg = MI.getOperand(0).getReg(); Src2Reg = MI.getOperand(2).getReg(); if (isIntRegForSubInst(Src1Reg) && isIntRegForSubInst(Src2Reg) && MI.getOperand(1).isImm() && isUInt<4>(MI.getOperand(1).getImm())) return HexagonII::HSIG_S1; break; // // Group S2: // // memh(Rs+#u3:1) = Rt // memw(r29+#u5:2) = Rt // memd(r29+#s6:3) = Rtt // memw(Rs+#u4:2) = #U1 // memb(Rs+#u4) = #U1 // allocframe(#u5:3) case Hexagon::S2_storerh_io: case Hexagon::dup_S2_storerh_io: // memh(Rs+#u3:1) = Rt Src1Reg = MI.getOperand(0).getReg(); Src2Reg = MI.getOperand(2).getReg(); if (isIntRegForSubInst(Src1Reg) && isIntRegForSubInst(Src2Reg) && MI.getOperand(1).isImm() && isShiftedUInt<3,1>(MI.getOperand(1).getImm())) return HexagonII::HSIG_S1; break; case Hexagon::S2_storerd_io: case Hexagon::dup_S2_storerd_io: // memd(r29+#s6:3) = Rtt Src1Reg = MI.getOperand(0).getReg(); Src2Reg = MI.getOperand(2).getReg(); if (isDblRegForSubInst(Src2Reg, HRI) && Hexagon::IntRegsRegClass.contains(Src1Reg) && HRI.getStackRegister() == Src1Reg && MI.getOperand(1).isImm() && isShiftedInt<6,3>(MI.getOperand(1).getImm())) return HexagonII::HSIG_S2; break; case Hexagon::S4_storeiri_io: case Hexagon::dup_S4_storeiri_io: // memw(Rs+#u4:2) = #U1 Src1Reg = MI.getOperand(0).getReg(); if (isIntRegForSubInst(Src1Reg) && MI.getOperand(1).isImm() && isShiftedUInt<4,2>(MI.getOperand(1).getImm()) && MI.getOperand(2).isImm() && isUInt<1>(MI.getOperand(2).getImm())) return HexagonII::HSIG_S2; break; case Hexagon::S4_storeirb_io: case Hexagon::dup_S4_storeirb_io: // memb(Rs+#u4) = #U1 Src1Reg = MI.getOperand(0).getReg(); if (isIntRegForSubInst(Src1Reg) && MI.getOperand(1).isImm() && isUInt<4>(MI.getOperand(1).getImm()) && MI.getOperand(2).isImm() && isUInt<1>(MI.getOperand(2).getImm())) return HexagonII::HSIG_S2; break; case Hexagon::S2_allocframe: case Hexagon::dup_S2_allocframe: if (MI.getOperand(2).isImm() && isShiftedUInt<5,3>(MI.getOperand(2).getImm())) return HexagonII::HSIG_S1; break; // // Group A: // // Rx = add(Rx,#s7) // Rd = Rs // Rd = #u6 // Rd = #-1 // if ([!]P0[.new]) Rd = #0 // Rd = add(r29,#u6:2) // Rx = add(Rx,Rs) // P0 = cmp.eq(Rs,#u2) // Rdd = combine(#0,Rs) // Rdd = combine(Rs,#0) // Rdd = combine(#u2,#U2) // Rd = add(Rs,#1) // Rd = add(Rs,#-1) // Rd = sxth/sxtb/zxtb/zxth(Rs) // Rd = and(Rs,#1) case Hexagon::A2_addi: case Hexagon::dup_A2_addi: DstReg = MI.getOperand(0).getReg(); SrcReg = MI.getOperand(1).getReg(); if (isIntRegForSubInst(DstReg)) { // Rd = add(r29,#u6:2) if (Hexagon::IntRegsRegClass.contains(SrcReg) && HRI.getStackRegister() == SrcReg && MI.getOperand(2).isImm() && isShiftedUInt<6,2>(MI.getOperand(2).getImm())) return HexagonII::HSIG_A; // Rx = add(Rx,#s7) if ((DstReg == SrcReg) && MI.getOperand(2).isImm() && isInt<7>(MI.getOperand(2).getImm())) return HexagonII::HSIG_A; // Rd = add(Rs,#1) // Rd = add(Rs,#-1) if (isIntRegForSubInst(SrcReg) && MI.getOperand(2).isImm() && ((MI.getOperand(2).getImm() == 1) || (MI.getOperand(2).getImm() == -1))) return HexagonII::HSIG_A; } break; case Hexagon::A2_add: case Hexagon::dup_A2_add: // Rx = add(Rx,Rs) DstReg = MI.getOperand(0).getReg(); Src1Reg = MI.getOperand(1).getReg(); Src2Reg = MI.getOperand(2).getReg(); if (isIntRegForSubInst(DstReg) && (DstReg == Src1Reg) && isIntRegForSubInst(Src2Reg)) return HexagonII::HSIG_A; break; case Hexagon::A2_andir: case Hexagon::dup_A2_andir: // Same as zxtb. // Rd16=and(Rs16,#255) // Rd16=and(Rs16,#1) DstReg = MI.getOperand(0).getReg(); SrcReg = MI.getOperand(1).getReg(); if (isIntRegForSubInst(DstReg) && isIntRegForSubInst(SrcReg) && MI.getOperand(2).isImm() && ((MI.getOperand(2).getImm() == 1) || (MI.getOperand(2).getImm() == 255))) return HexagonII::HSIG_A; break; case Hexagon::A2_tfr: case Hexagon::dup_A2_tfr: // Rd = Rs DstReg = MI.getOperand(0).getReg(); SrcReg = MI.getOperand(1).getReg(); if (isIntRegForSubInst(DstReg) && isIntRegForSubInst(SrcReg)) return HexagonII::HSIG_A; break; case Hexagon::A2_tfrsi: case Hexagon::dup_A2_tfrsi: // Rd = #u6 // Do not test for #u6 size since the const is getting extended // regardless and compound could be formed. // Rd = #-1 DstReg = MI.getOperand(0).getReg(); if (isIntRegForSubInst(DstReg)) return HexagonII::HSIG_A; break; case Hexagon::C2_cmoveit: case Hexagon::C2_cmovenewit: case Hexagon::C2_cmoveif: case Hexagon::C2_cmovenewif: case Hexagon::dup_C2_cmoveit: case Hexagon::dup_C2_cmovenewit: case Hexagon::dup_C2_cmoveif: case Hexagon::dup_C2_cmovenewif: // if ([!]P0[.new]) Rd = #0 // Actual form: // %r16 = C2_cmovenewit internal %p0, 0, implicit undef %r16; DstReg = MI.getOperand(0).getReg(); SrcReg = MI.getOperand(1).getReg(); if (isIntRegForSubInst(DstReg) && Hexagon::PredRegsRegClass.contains(SrcReg) && Hexagon::P0 == SrcReg && MI.getOperand(2).isImm() && MI.getOperand(2).getImm() == 0) return HexagonII::HSIG_A; break; case Hexagon::C2_cmpeqi: case Hexagon::dup_C2_cmpeqi: // P0 = cmp.eq(Rs,#u2) DstReg = MI.getOperand(0).getReg(); SrcReg = MI.getOperand(1).getReg(); if (Hexagon::PredRegsRegClass.contains(DstReg) && Hexagon::P0 == DstReg && isIntRegForSubInst(SrcReg) && MI.getOperand(2).isImm() && isUInt<2>(MI.getOperand(2).getImm())) return HexagonII::HSIG_A; break; case Hexagon::A2_combineii: case Hexagon::A4_combineii: case Hexagon::dup_A2_combineii: case Hexagon::dup_A4_combineii: // Rdd = combine(#u2,#U2) DstReg = MI.getOperand(0).getReg(); if (isDblRegForSubInst(DstReg, HRI) && ((MI.getOperand(1).isImm() && isUInt<2>(MI.getOperand(1).getImm())) || (MI.getOperand(1).isGlobal() && isUInt<2>(MI.getOperand(1).getOffset()))) && ((MI.getOperand(2).isImm() && isUInt<2>(MI.getOperand(2).getImm())) || (MI.getOperand(2).isGlobal() && isUInt<2>(MI.getOperand(2).getOffset())))) return HexagonII::HSIG_A; break; case Hexagon::A4_combineri: case Hexagon::dup_A4_combineri: // Rdd = combine(Rs,#0) // Rdd = combine(Rs,#0) DstReg = MI.getOperand(0).getReg(); SrcReg = MI.getOperand(1).getReg(); if (isDblRegForSubInst(DstReg, HRI) && isIntRegForSubInst(SrcReg) && ((MI.getOperand(2).isImm() && MI.getOperand(2).getImm() == 0) || (MI.getOperand(2).isGlobal() && MI.getOperand(2).getOffset() == 0))) return HexagonII::HSIG_A; break; case Hexagon::A4_combineir: case Hexagon::dup_A4_combineir: // Rdd = combine(#0,Rs) DstReg = MI.getOperand(0).getReg(); SrcReg = MI.getOperand(2).getReg(); if (isDblRegForSubInst(DstReg, HRI) && isIntRegForSubInst(SrcReg) && ((MI.getOperand(1).isImm() && MI.getOperand(1).getImm() == 0) || (MI.getOperand(1).isGlobal() && MI.getOperand(1).getOffset() == 0))) return HexagonII::HSIG_A; break; case Hexagon::A2_sxtb: case Hexagon::A2_sxth: case Hexagon::A2_zxtb: case Hexagon::A2_zxth: case Hexagon::dup_A2_sxtb: case Hexagon::dup_A2_sxth: case Hexagon::dup_A2_zxtb: case Hexagon::dup_A2_zxth: // Rd = sxth/sxtb/zxtb/zxth(Rs) DstReg = MI.getOperand(0).getReg(); SrcReg = MI.getOperand(1).getReg(); if (isIntRegForSubInst(DstReg) && isIntRegForSubInst(SrcReg)) return HexagonII::HSIG_A; break; } return HexagonII::HSIG_None; } short HexagonInstrInfo::getEquivalentHWInstr(const MachineInstr &MI) const { return Hexagon::getRealHWInstr(MI.getOpcode(), Hexagon::InstrType_Real); } unsigned HexagonInstrInfo::getInstrTimingClassLatency( const InstrItineraryData *ItinData, const MachineInstr &MI) const { // Default to one cycle for no itinerary. However, an "empty" itinerary may // still have a MinLatency property, which getStageLatency checks. if (!ItinData) return getInstrLatency(ItinData, MI); if (MI.isTransient()) return 0; return ItinData->getStageLatency(MI.getDesc().getSchedClass()); } /// getOperandLatency - Compute and return the use operand latency of a given /// pair of def and use. /// In most cases, the static scheduling itinerary was enough to determine the /// operand latency. But it may not be possible for instructions with variable /// number of defs / uses. /// /// This is a raw interface to the itinerary that may be directly overriden by /// a target. Use computeOperandLatency to get the best estimate of latency. int HexagonInstrInfo::getOperandLatency(const InstrItineraryData *ItinData, const MachineInstr &DefMI, unsigned DefIdx, const MachineInstr &UseMI, unsigned UseIdx) const { const HexagonRegisterInfo &HRI = *Subtarget.getRegisterInfo(); // Get DefIdx and UseIdx for super registers. const MachineOperand &DefMO = DefMI.getOperand(DefIdx); if (DefMO.isReg() && Register::isPhysicalRegister(DefMO.getReg())) { if (DefMO.isImplicit()) { for (MCSuperRegIterator SR(DefMO.getReg(), &HRI); SR.isValid(); ++SR) { int Idx = DefMI.findRegisterDefOperandIdx(*SR, false, false, &HRI); if (Idx != -1) { DefIdx = Idx; break; } } } const MachineOperand &UseMO = UseMI.getOperand(UseIdx); if (UseMO.isImplicit()) { for (MCSuperRegIterator SR(UseMO.getReg(), &HRI); SR.isValid(); ++SR) { int Idx = UseMI.findRegisterUseOperandIdx(*SR, false, &HRI); if (Idx != -1) { UseIdx = Idx; break; } } } } int Latency = TargetInstrInfo::getOperandLatency(ItinData, DefMI, DefIdx, UseMI, UseIdx); if (!Latency) // We should never have 0 cycle latency between two instructions unless // they can be packetized together. However, this decision can't be made // here. Latency = 1; return Latency; } // inverts the predication logic. // p -> NotP // NotP -> P bool HexagonInstrInfo::getInvertedPredSense( SmallVectorImpl &Cond) const { if (Cond.empty()) return false; unsigned Opc = getInvertedPredicatedOpcode(Cond[0].getImm()); Cond[0].setImm(Opc); return true; } unsigned HexagonInstrInfo::getInvertedPredicatedOpcode(const int Opc) const { int InvPredOpcode; InvPredOpcode = isPredicatedTrue(Opc) ? Hexagon::getFalsePredOpcode(Opc) : Hexagon::getTruePredOpcode(Opc); if (InvPredOpcode >= 0) // Valid instruction with the inverted predicate. return InvPredOpcode; llvm_unreachable("Unexpected predicated instruction"); } // Returns the max value that doesn't need to be extended. int HexagonInstrInfo::getMaxValue(const MachineInstr &MI) const { const uint64_t F = MI.getDesc().TSFlags; unsigned isSigned = (F >> HexagonII::ExtentSignedPos) & HexagonII::ExtentSignedMask; unsigned bits = (F >> HexagonII::ExtentBitsPos) & HexagonII::ExtentBitsMask; if (isSigned) // if value is signed return ~(-1U << (bits - 1)); else return ~(-1U << bits); } bool HexagonInstrInfo::isAddrModeWithOffset(const MachineInstr &MI) const { switch (MI.getOpcode()) { case Hexagon::L2_loadrbgp: case Hexagon::L2_loadrdgp: case Hexagon::L2_loadrhgp: case Hexagon::L2_loadrigp: case Hexagon::L2_loadrubgp: case Hexagon::L2_loadruhgp: case Hexagon::S2_storerbgp: case Hexagon::S2_storerbnewgp: case Hexagon::S2_storerhgp: case Hexagon::S2_storerhnewgp: case Hexagon::S2_storerigp: case Hexagon::S2_storerinewgp: case Hexagon::S2_storerdgp: case Hexagon::S2_storerfgp: return true; } const uint64_t F = MI.getDesc().TSFlags; unsigned addrMode = ((F >> HexagonII::AddrModePos) & HexagonII::AddrModeMask); // Disallow any base+offset instruction. The assembler does not yet reorder // based up any zero offset instruction. return (addrMode == HexagonII::BaseRegOffset || addrMode == HexagonII::BaseImmOffset || addrMode == HexagonII::BaseLongOffset); } bool HexagonInstrInfo::isPureSlot0(const MachineInstr &MI) const { // Workaround for the Global Scheduler. Sometimes, it creates // A4_ext as a Pseudo instruction and calls this function to see if // it can be added to an existing bundle. Since the instruction doesn't // belong to any BB yet, we can't use getUnits API. if (MI.getOpcode() == Hexagon::A4_ext) return false; unsigned FuncUnits = getUnits(MI); return HexagonFUnits::isSlot0Only(FuncUnits); } bool HexagonInstrInfo::isRestrictNoSlot1Store(const MachineInstr &MI) const { const uint64_t F = MI.getDesc().TSFlags; return ((F >> HexagonII::RestrictNoSlot1StorePos) & HexagonII::RestrictNoSlot1StoreMask); } void HexagonInstrInfo::changeDuplexOpcode(MachineBasicBlock::instr_iterator MII, bool ToBigInstrs) const { int Opcode = -1; if (ToBigInstrs) { // To BigCore Instr. // Check if the instruction can form a Duplex. if (getDuplexCandidateGroup(*MII)) // Get the opcode marked "dup_*" tag. Opcode = getDuplexOpcode(*MII, ToBigInstrs); } else // To TinyCore Instr. Opcode = getDuplexOpcode(*MII, ToBigInstrs); // Change the opcode of the instruction. if (Opcode >= 0) MII->setDesc(get(Opcode)); } // This function is used to translate instructions to facilitate generating // Duplexes on TinyCore. void HexagonInstrInfo::translateInstrsForDup(MachineFunction &MF, bool ToBigInstrs) const { for (auto &MB : MF) for (MachineBasicBlock::instr_iterator Instr = MB.instr_begin(), End = MB.instr_end(); Instr != End; ++Instr) changeDuplexOpcode(Instr, ToBigInstrs); } // This is a specialized form of above function. void HexagonInstrInfo::translateInstrsForDup( MachineBasicBlock::instr_iterator MII, bool ToBigInstrs) const { MachineBasicBlock *MBB = MII->getParent(); while ((MII != MBB->instr_end()) && MII->isInsideBundle()) { changeDuplexOpcode(MII, ToBigInstrs); ++MII; } } unsigned HexagonInstrInfo::getMemAccessSize(const MachineInstr &MI) const { using namespace HexagonII; const uint64_t F = MI.getDesc().TSFlags; unsigned S = (F >> MemAccessSizePos) & MemAccesSizeMask; unsigned Size = getMemAccessSizeInBytes(MemAccessSize(S)); if (Size != 0) return Size; // Handle vector access sizes. const HexagonRegisterInfo &HRI = *Subtarget.getRegisterInfo(); switch (S) { case HexagonII::HVXVectorAccess: return HRI.getSpillSize(Hexagon::HvxVRRegClass); default: llvm_unreachable("Unexpected instruction"); } } // Returns the min value that doesn't need to be extended. int HexagonInstrInfo::getMinValue(const MachineInstr &MI) const { const uint64_t F = MI.getDesc().TSFlags; unsigned isSigned = (F >> HexagonII::ExtentSignedPos) & HexagonII::ExtentSignedMask; unsigned bits = (F >> HexagonII::ExtentBitsPos) & HexagonII::ExtentBitsMask; if (isSigned) // if value is signed return -1U << (bits - 1); else return 0; } // Returns opcode of the non-extended equivalent instruction. short HexagonInstrInfo::getNonExtOpcode(const MachineInstr &MI) const { // Check if the instruction has a register form that uses register in place // of the extended operand, if so return that as the non-extended form. short NonExtOpcode = Hexagon::getRegForm(MI.getOpcode()); if (NonExtOpcode >= 0) return NonExtOpcode; if (MI.getDesc().mayLoad() || MI.getDesc().mayStore()) { // Check addressing mode and retrieve non-ext equivalent instruction. switch (getAddrMode(MI)) { case HexagonII::Absolute: return Hexagon::changeAddrMode_abs_io(MI.getOpcode()); case HexagonII::BaseImmOffset: return Hexagon::changeAddrMode_io_rr(MI.getOpcode()); case HexagonII::BaseLongOffset: return Hexagon::changeAddrMode_ur_rr(MI.getOpcode()); default: return -1; } } return -1; } bool HexagonInstrInfo::getPredReg(ArrayRef Cond, unsigned &PredReg, unsigned &PredRegPos, unsigned &PredRegFlags) const { if (Cond.empty()) return false; assert(Cond.size() == 2); if (isNewValueJump(Cond[0].getImm()) || Cond[1].isMBB()) { LLVM_DEBUG(dbgs() << "No predregs for new-value jumps/endloop"); return false; } PredReg = Cond[1].getReg(); PredRegPos = 1; // See IfConversion.cpp why we add RegState::Implicit | RegState::Undef PredRegFlags = 0; if (Cond[1].isImplicit()) PredRegFlags = RegState::Implicit; if (Cond[1].isUndef()) PredRegFlags |= RegState::Undef; return true; } short HexagonInstrInfo::getPseudoInstrPair(const MachineInstr &MI) const { return Hexagon::getRealHWInstr(MI.getOpcode(), Hexagon::InstrType_Pseudo); } short HexagonInstrInfo::getRegForm(const MachineInstr &MI) const { return Hexagon::getRegForm(MI.getOpcode()); } // Return the number of bytes required to encode the instruction. // Hexagon instructions are fixed length, 4 bytes, unless they // use a constant extender, which requires another 4 bytes. // For debug instructions and prolog labels, return 0. unsigned HexagonInstrInfo::getSize(const MachineInstr &MI) const { if (MI.isDebugInstr() || MI.isPosition()) return 0; unsigned Size = MI.getDesc().getSize(); if (!Size) // Assume the default insn size in case it cannot be determined // for whatever reason. Size = HEXAGON_INSTR_SIZE; if (isConstExtended(MI) || isExtended(MI)) Size += HEXAGON_INSTR_SIZE; // Try and compute number of instructions in asm. if (BranchRelaxAsmLarge && MI.getOpcode() == Hexagon::INLINEASM) { const MachineBasicBlock &MBB = *MI.getParent(); const MachineFunction *MF = MBB.getParent(); const MCAsmInfo *MAI = MF->getTarget().getMCAsmInfo(); // Count the number of register definitions to find the asm string. unsigned NumDefs = 0; for (; MI.getOperand(NumDefs).isReg() && MI.getOperand(NumDefs).isDef(); ++NumDefs) assert(NumDefs != MI.getNumOperands()-2 && "No asm string?"); assert(MI.getOperand(NumDefs).isSymbol() && "No asm string?"); // Disassemble the AsmStr and approximate number of instructions. const char *AsmStr = MI.getOperand(NumDefs).getSymbolName(); Size = getInlineAsmLength(AsmStr, *MAI); } return Size; } uint64_t HexagonInstrInfo::getType(const MachineInstr &MI) const { const uint64_t F = MI.getDesc().TSFlags; return (F >> HexagonII::TypePos) & HexagonII::TypeMask; } InstrStage::FuncUnits HexagonInstrInfo::getUnits(const MachineInstr &MI) const { const InstrItineraryData &II = *Subtarget.getInstrItineraryData(); const InstrStage &IS = *II.beginStage(MI.getDesc().getSchedClass()); return IS.getUnits(); } // Calculate size of the basic block without debug instructions. unsigned HexagonInstrInfo::nonDbgBBSize(const MachineBasicBlock *BB) const { return nonDbgMICount(BB->instr_begin(), BB->instr_end()); } unsigned HexagonInstrInfo::nonDbgBundleSize( MachineBasicBlock::const_iterator BundleHead) const { assert(BundleHead->isBundle() && "Not a bundle header"); auto MII = BundleHead.getInstrIterator(); // Skip the bundle header. return nonDbgMICount(++MII, getBundleEnd(BundleHead.getInstrIterator())); } /// immediateExtend - Changes the instruction in place to one using an immediate /// extender. void HexagonInstrInfo::immediateExtend(MachineInstr &MI) const { assert((isExtendable(MI)||isConstExtended(MI)) && "Instruction must be extendable"); // Find which operand is extendable. short ExtOpNum = getCExtOpNum(MI); MachineOperand &MO = MI.getOperand(ExtOpNum); // This needs to be something we understand. assert((MO.isMBB() || MO.isImm()) && "Branch with unknown extendable field type"); // Mark given operand as extended. MO.addTargetFlag(HexagonII::HMOTF_ConstExtended); } bool HexagonInstrInfo::invertAndChangeJumpTarget( MachineInstr &MI, MachineBasicBlock *NewTarget) const { LLVM_DEBUG(dbgs() << "\n[invertAndChangeJumpTarget] to " << printMBBReference(*NewTarget); MI.dump();); assert(MI.isBranch()); unsigned NewOpcode = getInvertedPredicatedOpcode(MI.getOpcode()); int TargetPos = MI.getNumOperands() - 1; // In general branch target is the last operand, // but some implicit defs added at the end might change it. while ((TargetPos > -1) && !MI.getOperand(TargetPos).isMBB()) --TargetPos; assert((TargetPos >= 0) && MI.getOperand(TargetPos).isMBB()); MI.getOperand(TargetPos).setMBB(NewTarget); if (EnableBranchPrediction && isPredicatedNew(MI)) { NewOpcode = reversePrediction(NewOpcode); } MI.setDesc(get(NewOpcode)); return true; } void HexagonInstrInfo::genAllInsnTimingClasses(MachineFunction &MF) const { /* +++ The code below is used to generate complete set of Hexagon Insn +++ */ MachineFunction::iterator A = MF.begin(); MachineBasicBlock &B = *A; MachineBasicBlock::iterator I = B.begin(); DebugLoc DL = I->getDebugLoc(); MachineInstr *NewMI; for (unsigned insn = TargetOpcode::GENERIC_OP_END+1; insn < Hexagon::INSTRUCTION_LIST_END; ++insn) { NewMI = BuildMI(B, I, DL, get(insn)); LLVM_DEBUG(dbgs() << "\n" << getName(NewMI->getOpcode()) << " Class: " << NewMI->getDesc().getSchedClass()); NewMI->eraseFromParent(); } /* --- The code above is used to generate complete set of Hexagon Insn --- */ } // inverts the predication logic. // p -> NotP // NotP -> P bool HexagonInstrInfo::reversePredSense(MachineInstr &MI) const { LLVM_DEBUG(dbgs() << "\nTrying to reverse pred. sense of:"; MI.dump()); MI.setDesc(get(getInvertedPredicatedOpcode(MI.getOpcode()))); return true; } // Reverse the branch prediction. unsigned HexagonInstrInfo::reversePrediction(unsigned Opcode) const { int PredRevOpcode = -1; if (isPredictedTaken(Opcode)) PredRevOpcode = Hexagon::notTakenBranchPrediction(Opcode); else PredRevOpcode = Hexagon::takenBranchPrediction(Opcode); assert(PredRevOpcode > 0); return PredRevOpcode; } // TODO: Add more rigorous validation. bool HexagonInstrInfo::validateBranchCond(const ArrayRef &Cond) const { return Cond.empty() || (Cond[0].isImm() && (Cond.size() != 1)); } void HexagonInstrInfo:: setBundleNoShuf(MachineBasicBlock::instr_iterator MIB) const { assert(MIB->isBundle()); MachineOperand &Operand = MIB->getOperand(0); if (Operand.isImm()) Operand.setImm(Operand.getImm() | memShufDisabledMask); else MIB->addOperand(MachineOperand::CreateImm(memShufDisabledMask)); } bool HexagonInstrInfo::getBundleNoShuf(const MachineInstr &MIB) const { assert(MIB.isBundle()); const MachineOperand &Operand = MIB.getOperand(0); return (Operand.isImm() && (Operand.getImm() & memShufDisabledMask) != 0); } // Addressing mode relations. short HexagonInstrInfo::changeAddrMode_abs_io(short Opc) const { return Opc >= 0 ? Hexagon::changeAddrMode_abs_io(Opc) : Opc; } short HexagonInstrInfo::changeAddrMode_io_abs(short Opc) const { return Opc >= 0 ? Hexagon::changeAddrMode_io_abs(Opc) : Opc; } short HexagonInstrInfo::changeAddrMode_io_pi(short Opc) const { return Opc >= 0 ? Hexagon::changeAddrMode_io_pi(Opc) : Opc; } short HexagonInstrInfo::changeAddrMode_io_rr(short Opc) const { return Opc >= 0 ? Hexagon::changeAddrMode_io_rr(Opc) : Opc; } short HexagonInstrInfo::changeAddrMode_pi_io(short Opc) const { return Opc >= 0 ? Hexagon::changeAddrMode_pi_io(Opc) : Opc; } short HexagonInstrInfo::changeAddrMode_rr_io(short Opc) const { return Opc >= 0 ? Hexagon::changeAddrMode_rr_io(Opc) : Opc; } short HexagonInstrInfo::changeAddrMode_rr_ur(short Opc) const { return Opc >= 0 ? Hexagon::changeAddrMode_rr_ur(Opc) : Opc; } short HexagonInstrInfo::changeAddrMode_ur_rr(short Opc) const { return Opc >= 0 ? Hexagon::changeAddrMode_ur_rr(Opc) : Opc; }