brintos

brintos / llvm-project-archived public Read only

0
0
Text · 26.7 KiB · 05667f4 Raw
818 lines · cpp
1//===- ARMParallelDSP.cpp - Parallel DSP Pass -----------------------------===//2//3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.4// See https://llvm.org/LICENSE.txt for license information.5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception6//7//===----------------------------------------------------------------------===//8//9/// \file10/// Armv6 introduced instructions to perform 32-bit SIMD operations. The11/// purpose of this pass is do some IR pattern matching to create ACLE12/// DSP intrinsics, which map on these 32-bit SIMD operations.13/// This pass runs only when unaligned accesses is supported/enabled.14//15//===----------------------------------------------------------------------===//16 17#include "ARM.h"18#include "ARMSubtarget.h"19#include "llvm/ADT/SmallPtrSet.h"20#include "llvm/ADT/Statistic.h"21#include "llvm/Analysis/AliasAnalysis.h"22#include "llvm/Analysis/AssumptionCache.h"23#include "llvm/Analysis/GlobalsModRef.h"24#include "llvm/Analysis/LoopAccessAnalysis.h"25#include "llvm/Analysis/TargetLibraryInfo.h"26#include "llvm/CodeGen/TargetPassConfig.h"27#include "llvm/IR/IRBuilder.h"28#include "llvm/IR/Instructions.h"29#include "llvm/IR/IntrinsicsARM.h"30#include "llvm/IR/Module.h"31#include "llvm/IR/NoFolder.h"32#include "llvm/IR/PatternMatch.h"33#include "llvm/Pass.h"34#include "llvm/Support/Debug.h"35#include "llvm/Transforms/Scalar.h"36#include "llvm/Transforms/Utils/BasicBlockUtils.h"37 38using namespace llvm;39using namespace PatternMatch;40 41#define DEBUG_TYPE "arm-parallel-dsp"42 43STATISTIC(NumSMLAD , "Number of smlad instructions generated");44 45static cl::opt<bool>46DisableParallelDSP("disable-arm-parallel-dsp", cl::Hidden, cl::init(false),47                   cl::desc("Disable the ARM Parallel DSP pass"));48 49static cl::opt<unsigned>50NumLoadLimit("arm-parallel-dsp-load-limit", cl::Hidden, cl::init(16),51             cl::desc("Limit the number of loads analysed"));52 53namespace {54  struct MulCandidate;55  class Reduction;56 57  using MulCandList = SmallVector<std::unique_ptr<MulCandidate>, 8>;58  using MemInstList = SmallVectorImpl<LoadInst*>;59  using MulPairList = SmallVector<std::pair<MulCandidate*, MulCandidate*>, 8>;60 61  // 'MulCandidate' holds the multiplication instructions that are candidates62  // for parallel execution.63  struct MulCandidate {64    Instruction   *Root;65    Value*        LHS;66    Value*        RHS;67    bool          Exchange = false;68    bool          Paired = false;69    SmallVector<LoadInst*, 2> VecLd;    // Container for loads to widen.70 71    MulCandidate(Instruction *I, Value *lhs, Value *rhs) :72      Root(I), LHS(lhs), RHS(rhs) { }73 74    bool HasTwoLoadInputs() const {75      return isa<LoadInst>(LHS) && isa<LoadInst>(RHS);76    }77 78    LoadInst *getBaseLoad() const {79      return VecLd.front();80    }81  };82 83  /// Represent a sequence of multiply-accumulate operations with the aim to84  /// perform the multiplications in parallel.85  class Reduction {86    Instruction     *Root = nullptr;87    Value           *Acc = nullptr;88    MulCandList     Muls;89    MulPairList        MulPairs;90    SetVector<Instruction*> Adds;91 92  public:93    Reduction() = delete;94 95    Reduction (Instruction *Add) : Root(Add) { }96 97    /// Record an Add instruction that is a part of the this reduction.98    void InsertAdd(Instruction *I) { Adds.insert(I); }99 100    /// Create MulCandidates, each rooted at a Mul instruction, that is a part101    /// of this reduction.102    void InsertMuls() {103      auto GetMulOperand = [](Value *V) -> Instruction* {104        if (auto *SExt = dyn_cast<SExtInst>(V)) {105          if (auto *I = dyn_cast<Instruction>(SExt->getOperand(0)))106            if (I->getOpcode() == Instruction::Mul)107              return I;108        } else if (auto *I = dyn_cast<Instruction>(V)) {109          if (I->getOpcode() == Instruction::Mul)110            return I;111        }112        return nullptr;113      };114 115      auto InsertMul = [this](Instruction *I) {116        Value *LHS = cast<Instruction>(I->getOperand(0))->getOperand(0);117        Value *RHS = cast<Instruction>(I->getOperand(1))->getOperand(0);118        Muls.push_back(std::make_unique<MulCandidate>(I, LHS, RHS));119      };120 121      for (auto *Add : Adds) {122        if (Add == Acc)123          continue;124        if (auto *Mul = GetMulOperand(Add->getOperand(0)))125          InsertMul(Mul);126        if (auto *Mul = GetMulOperand(Add->getOperand(1)))127          InsertMul(Mul);128      }129    }130 131    /// Add the incoming accumulator value, returns true if a value had not132    /// already been added. Returning false signals to the user that this133    /// reduction already has a value to initialise the accumulator.134    bool InsertAcc(Value *V) {135      if (Acc)136        return false;137      Acc = V;138      return true;139    }140 141    /// Set two MulCandidates, rooted at muls, that can be executed as a single142    /// parallel operation.143    void AddMulPair(MulCandidate *Mul0, MulCandidate *Mul1,144                    bool Exchange = false) {145      LLVM_DEBUG(dbgs() << "Pairing:\n"146                 << *Mul0->Root << "\n"147                 << *Mul1->Root << "\n");148      Mul0->Paired = true;149      Mul1->Paired = true;150      if (Exchange)151        Mul1->Exchange = true;152      MulPairs.push_back(std::make_pair(Mul0, Mul1));153    }154 155    /// Return the add instruction which is the root of the reduction.156    Instruction *getRoot() { return Root; }157 158    bool is64Bit() const { return Root->getType()->isIntegerTy(64); }159 160    Type *getType() const { return Root->getType(); }161 162    /// Return the incoming value to be accumulated. This maybe null.163    Value *getAccumulator() { return Acc; }164 165    /// Return the set of adds that comprise the reduction.166    SetVector<Instruction*> &getAdds() { return Adds; }167 168    /// Return the MulCandidate, rooted at mul instruction, that comprise the169    /// the reduction.170    MulCandList &getMuls() { return Muls; }171 172    /// Return the MulCandidate, rooted at mul instructions, that have been173    /// paired for parallel execution.174    MulPairList &getMulPairs() { return MulPairs; }175 176    /// To finalise, replace the uses of the root with the intrinsic call.177    void UpdateRoot(Instruction *SMLAD) {178      Root->replaceAllUsesWith(SMLAD);179    }180 181    void dump() {182      LLVM_DEBUG(dbgs() << "Reduction:\n";183        for (auto *Add : Adds)184          LLVM_DEBUG(dbgs() << *Add << "\n");185        for (auto &Mul : Muls)186          LLVM_DEBUG(dbgs() << *Mul->Root << "\n"187                     << "  " << *Mul->LHS << "\n"188                     << "  " << *Mul->RHS << "\n");189        LLVM_DEBUG(if (Acc) dbgs() << "Acc in: " << *Acc << "\n")190      );191    }192  };193 194  class WidenedLoad {195    LoadInst *NewLd = nullptr;196    SmallVector<LoadInst*, 4> Loads;197 198  public:199    WidenedLoad(SmallVectorImpl<LoadInst*> &Lds, LoadInst *Wide)200      : NewLd(Wide) {201      append_range(Loads, Lds);202    }203    LoadInst *getLoad() {204      return NewLd;205    }206  };207 208  class ARMParallelDSP : public FunctionPass {209    ScalarEvolution   *SE;210    AliasAnalysis     *AA;211    TargetLibraryInfo *TLI;212    DominatorTree     *DT;213    const DataLayout  *DL;214    Module            *M;215    std::map<LoadInst*, LoadInst*> LoadPairs;216    SmallPtrSet<LoadInst*, 4> OffsetLoads;217    std::map<LoadInst*, std::unique_ptr<WidenedLoad>> WideLoads;218 219    template<unsigned>220    bool IsNarrowSequence(Value *V);221    bool Search(Value *V, BasicBlock *BB, Reduction &R);222    bool RecordMemoryOps(BasicBlock *BB);223    void InsertParallelMACs(Reduction &Reduction);224    bool AreSequentialLoads(LoadInst *Ld0, LoadInst *Ld1, MemInstList &VecMem);225    LoadInst* CreateWideLoad(MemInstList &Loads, IntegerType *LoadTy);226    bool CreateParallelPairs(Reduction &R);227 228    /// Try to match and generate: SMLAD, SMLADX - Signed Multiply Accumulate229    /// Dual performs two signed 16x16-bit multiplications. It adds the230    /// products to a 32-bit accumulate operand. Optionally, the instruction can231    /// exchange the halfwords of the second operand before performing the232    /// arithmetic.233    bool MatchSMLAD(Function &F);234 235  public:236    static char ID;237 238    ARMParallelDSP() : FunctionPass(ID) { }239 240    void getAnalysisUsage(AnalysisUsage &AU) const override {241      FunctionPass::getAnalysisUsage(AU);242      AU.addRequired<AssumptionCacheTracker>();243      AU.addRequired<ScalarEvolutionWrapperPass>();244      AU.addRequired<AAResultsWrapperPass>();245      AU.addRequired<TargetLibraryInfoWrapperPass>();246      AU.addRequired<DominatorTreeWrapperPass>();247      AU.addRequired<TargetPassConfig>();248      AU.addPreserved<ScalarEvolutionWrapperPass>();249      AU.addPreserved<GlobalsAAWrapperPass>();250      AU.setPreservesCFG();251    }252 253    bool runOnFunction(Function &F) override {254      if (DisableParallelDSP)255        return false;256      if (skipFunction(F))257        return false;258 259      SE = &getAnalysis<ScalarEvolutionWrapperPass>().getSE();260      AA = &getAnalysis<AAResultsWrapperPass>().getAAResults();261      TLI = &getAnalysis<TargetLibraryInfoWrapperPass>().getTLI(F);262      DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree();263      auto &TPC = getAnalysis<TargetPassConfig>();264 265      M = F.getParent();266      DL = &M->getDataLayout();267 268      auto &TM = TPC.getTM<TargetMachine>();269      auto *ST = &TM.getSubtarget<ARMSubtarget>(F);270 271      if (!ST->allowsUnalignedMem()) {272        LLVM_DEBUG(dbgs() << "Unaligned memory access not supported: not "273                             "running pass ARMParallelDSP\n");274        return false;275      }276 277      if (!ST->hasDSP()) {278        LLVM_DEBUG(dbgs() << "DSP extension not enabled: not running pass "279                             "ARMParallelDSP\n");280        return false;281      }282 283      if (!ST->isLittle()) {284        LLVM_DEBUG(dbgs() << "Only supporting little endian: not running pass "285                          << "ARMParallelDSP\n");286        return false;287      }288 289      LLVM_DEBUG(dbgs() << "\n== Parallel DSP pass ==\n");290      LLVM_DEBUG(dbgs() << " - " << F.getName() << "\n\n");291 292      bool Changes = MatchSMLAD(F);293      return Changes;294    }295  };296}297 298bool ARMParallelDSP::AreSequentialLoads(LoadInst *Ld0, LoadInst *Ld1,299                                        MemInstList &VecMem) {300  if (!Ld0 || !Ld1)301    return false;302 303  auto It = LoadPairs.find(Ld0);304  if (It == LoadPairs.end() || It->second != Ld1)305    return false;306 307  LLVM_DEBUG(dbgs() << "Loads are sequential and valid:\n";308    dbgs() << "Ld0:"; Ld0->dump();309    dbgs() << "Ld1:"; Ld1->dump();310  );311 312  VecMem.clear();313  VecMem.push_back(Ld0);314  VecMem.push_back(Ld1);315  return true;316}317 318// MaxBitwidth: the maximum supported bitwidth of the elements in the DSP319// instructions, which is set to 16. So here we should collect all i8 and i16320// narrow operations.321// TODO: we currently only collect i16, and will support i8 later, so that's322// why we check that types are equal to MaxBitWidth, and not <= MaxBitWidth.323template<unsigned MaxBitWidth>324bool ARMParallelDSP::IsNarrowSequence(Value *V) {325  if (auto *SExt = dyn_cast<SExtInst>(V)) {326    if (SExt->getSrcTy()->getIntegerBitWidth() != MaxBitWidth)327      return false;328 329    if (auto *Ld = dyn_cast<LoadInst>(SExt->getOperand(0))) {330      // Check that this load could be paired.331      return LoadPairs.count(Ld) || OffsetLoads.count(Ld);332    }333  }334  return false;335}336 337/// Iterate through the block and record base, offset pairs of loads which can338/// be widened into a single load.339bool ARMParallelDSP::RecordMemoryOps(BasicBlock *BB) {340  SmallVector<LoadInst*, 8> Loads;341  SmallVector<Instruction*, 8> Writes;342  LoadPairs.clear();343  WideLoads.clear();344 345  // Collect loads and instruction that may write to memory. For now we only346  // record loads which are simple, sign-extended and have a single user.347  // TODO: Allow zero-extended loads.348  for (auto &I : *BB) {349    if (I.mayWriteToMemory())350      Writes.push_back(&I);351    auto *Ld = dyn_cast<LoadInst>(&I);352    if (!Ld || !Ld->isSimple() ||353        !Ld->hasOneUse() || !isa<SExtInst>(Ld->user_back()))354      continue;355    Loads.push_back(Ld);356  }357 358  if (Loads.empty() || Loads.size() > NumLoadLimit)359    return false;360 361  using InstSet = std::set<Instruction*>;362  using DepMap = std::map<Instruction*, InstSet>;363  DepMap RAWDeps;364 365  // Record any writes that may alias a load.366  const auto Size = LocationSize::beforeOrAfterPointer();367  for (auto *Write : Writes) {368    for (auto *Read : Loads) {369      MemoryLocation ReadLoc =370        MemoryLocation(Read->getPointerOperand(), Size);371 372      if (!isModOrRefSet(AA->getModRefInfo(Write, ReadLoc)))373        continue;374      if (Write->comesBefore(Read))375        RAWDeps[Read].insert(Write);376    }377  }378 379  // Check whether there's not a write between the two loads which would380  // prevent them from being safely merged.381  auto SafeToPair = [&](LoadInst *Base, LoadInst *Offset) {382    bool BaseFirst = Base->comesBefore(Offset);383    LoadInst *Dominator = BaseFirst ? Base : Offset;384    LoadInst *Dominated = BaseFirst ? Offset : Base;385 386    if (auto It = RAWDeps.find(Dominated); It != RAWDeps.end()) {387      InstSet &WritesBefore = It->second;388 389      for (auto *Before : WritesBefore) {390        // We can't move the second load backward, past a write, to merge391        // with the first load.392        if (Dominator->comesBefore(Before))393          return false;394      }395    }396    return true;397  };398 399  // Record base, offset load pairs.400  for (auto *Base : Loads) {401    for (auto *Offset : Loads) {402      if (Base == Offset || OffsetLoads.count(Offset))403        continue;404 405      if (isConsecutiveAccess(Base, Offset, *DL, *SE) &&406          SafeToPair(Base, Offset)) {407        LoadPairs[Base] = Offset;408        OffsetLoads.insert(Offset);409        break;410      }411    }412  }413 414  LLVM_DEBUG(if (!LoadPairs.empty()) {415               dbgs() << "Consecutive load pairs:\n";416               for (auto &MapIt : LoadPairs) {417                 LLVM_DEBUG(dbgs() << *MapIt.first << ", "418                            << *MapIt.second << "\n");419               }420             });421  return LoadPairs.size() > 1;422}423 424// Search recursively back through the operands to find a tree of values that425// form a multiply-accumulate chain. The search records the Add and Mul426// instructions that form the reduction and allows us to find a single value427// to be used as the initial input to the accumlator.428bool ARMParallelDSP::Search(Value *V, BasicBlock *BB, Reduction &R) {429  // If we find a non-instruction, try to use it as the initial accumulator430  // value. This may have already been found during the search in which case431  // this function will return false, signaling a search fail.432  auto *I = dyn_cast<Instruction>(V);433  if (!I)434    return R.InsertAcc(V);435 436  if (I->getParent() != BB)437    return false;438 439  switch (I->getOpcode()) {440  default:441    break;442  case Instruction::PHI:443    // Could be the accumulator value.444    return R.InsertAcc(V);445  case Instruction::Add: {446    // Adds should be adding together two muls, or another add and a mul to447    // be within the mac chain. One of the operands may also be the448    // accumulator value at which point we should stop searching.449    R.InsertAdd(I);450    Value *LHS = I->getOperand(0);451    Value *RHS = I->getOperand(1);452    bool ValidLHS = Search(LHS, BB, R);453    bool ValidRHS = Search(RHS, BB, R);454 455    if (ValidLHS && ValidRHS)456      return true;457 458    // Ensure we don't add the root as the incoming accumulator.459    if (R.getRoot() == I)460      return false;461 462    return R.InsertAcc(I);463  }464  case Instruction::Mul: {465    Value *MulOp0 = I->getOperand(0);466    Value *MulOp1 = I->getOperand(1);467    return IsNarrowSequence<16>(MulOp0) && IsNarrowSequence<16>(MulOp1);468  }469  case Instruction::SExt:470    return Search(I->getOperand(0), BB, R);471  }472  return false;473}474 475// The pass needs to identify integer add/sub reductions of 16-bit vector476// multiplications.477// To use SMLAD:478// 1) we first need to find integer add then look for this pattern:479//480// acc0 = ...481// ld0 = load i16482// sext0 = sext i16 %ld0 to i32483// ld1 = load i16484// sext1 = sext i16 %ld1 to i32485// mul0 = mul %sext0, %sext1486// ld2 = load i16487// sext2 = sext i16 %ld2 to i32488// ld3 = load i16489// sext3 = sext i16 %ld3 to i32490// mul1 = mul i32 %sext2, %sext3491// add0 = add i32 %mul0, %acc0492// acc1 = add i32 %add0, %mul1493//494// Which can be selected to:495//496// ldr r0497// ldr r1498// smlad r2, r0, r1, r2499//500// If constants are used instead of loads, these will need to be hoisted501// out and into a register.502//503// If loop invariants are used instead of loads, these need to be packed504// before the loop begins.505//506bool ARMParallelDSP::MatchSMLAD(Function &F) {507  bool Changed = false;508 509  for (auto &BB : F) {510    SmallPtrSet<Instruction*, 4> AllAdds;511    if (!RecordMemoryOps(&BB))512      continue;513 514    for (Instruction &I : reverse(BB)) {515      if (I.getOpcode() != Instruction::Add)516        continue;517 518      if (AllAdds.count(&I))519        continue;520 521      const auto *Ty = I.getType();522      if (!Ty->isIntegerTy(32) && !Ty->isIntegerTy(64))523        continue;524 525      Reduction R(&I);526      if (!Search(&I, &BB, R))527        continue;528 529      R.InsertMuls();530      LLVM_DEBUG(dbgs() << "After search, Reduction:\n"; R.dump());531 532      if (!CreateParallelPairs(R))533        continue;534 535      InsertParallelMACs(R);536      Changed = true;537      AllAdds.insert_range(R.getAdds());538      LLVM_DEBUG(dbgs() << "BB after inserting parallel MACs:\n" << BB);539    }540  }541 542  return Changed;543}544 545bool ARMParallelDSP::CreateParallelPairs(Reduction &R) {546 547  // Not enough mul operations to make a pair.548  if (R.getMuls().size() < 2)549    return false;550 551  // Check that the muls operate directly upon sign extended loads.552  for (auto &MulCand : R.getMuls()) {553    if (!MulCand->HasTwoLoadInputs())554      return false;555  }556 557  auto CanPair = [&](Reduction &R, MulCandidate *PMul0, MulCandidate *PMul1) {558    // The first elements of each vector should be loads with sexts. If we559    // find that its two pairs of consecutive loads, then these can be560    // transformed into two wider loads and the users can be replaced with561    // DSP intrinsics.562    auto Ld0 = static_cast<LoadInst*>(PMul0->LHS);563    auto Ld1 = static_cast<LoadInst*>(PMul1->LHS);564    auto Ld2 = static_cast<LoadInst*>(PMul0->RHS);565    auto Ld3 = static_cast<LoadInst*>(PMul1->RHS);566 567    // Check that each mul is operating on two different loads.568    if (Ld0 == Ld2 || Ld1 == Ld3)569      return false;570 571    if (AreSequentialLoads(Ld0, Ld1, PMul0->VecLd)) {572      if (AreSequentialLoads(Ld2, Ld3, PMul1->VecLd)) {573        LLVM_DEBUG(dbgs() << "OK: found two pairs of parallel loads!\n");574        R.AddMulPair(PMul0, PMul1);575        return true;576      } else if (AreSequentialLoads(Ld3, Ld2, PMul1->VecLd)) {577        LLVM_DEBUG(dbgs() << "OK: found two pairs of parallel loads!\n");578        LLVM_DEBUG(dbgs() << "    exchanging Ld2 and Ld3\n");579        R.AddMulPair(PMul0, PMul1, true);580        return true;581      }582    } else if (AreSequentialLoads(Ld1, Ld0, PMul0->VecLd) &&583               AreSequentialLoads(Ld2, Ld3, PMul1->VecLd)) {584      LLVM_DEBUG(dbgs() << "OK: found two pairs of parallel loads!\n");585      LLVM_DEBUG(dbgs() << "    exchanging Ld0 and Ld1\n");586      LLVM_DEBUG(dbgs() << "    and swapping muls\n");587      // Only the second operand can be exchanged, so swap the muls.588      R.AddMulPair(PMul1, PMul0, true);589      return true;590    }591    return false;592  };593 594  MulCandList &Muls = R.getMuls();595  const unsigned Elems = Muls.size();596  for (unsigned i = 0; i < Elems; ++i) {597    MulCandidate *PMul0 = static_cast<MulCandidate*>(Muls[i].get());598    if (PMul0->Paired)599      continue;600 601    for (unsigned j = 0; j < Elems; ++j) {602      if (i == j)603        continue;604 605      MulCandidate *PMul1 = static_cast<MulCandidate*>(Muls[j].get());606      if (PMul1->Paired)607        continue;608 609      const Instruction *Mul0 = PMul0->Root;610      const Instruction *Mul1 = PMul1->Root;611      if (Mul0 == Mul1)612        continue;613 614      assert(PMul0 != PMul1 && "expected different chains");615 616      if (CanPair(R, PMul0, PMul1))617        break;618    }619  }620  return !R.getMulPairs().empty();621}622 623void ARMParallelDSP::InsertParallelMACs(Reduction &R) {624 625  auto CreateSMLAD = [&](LoadInst* WideLd0, LoadInst *WideLd1,626                         Value *Acc, bool Exchange,627                         Instruction *InsertAfter) {628    // Replace the reduction chain with an intrinsic call629 630    Value* Args[] = { WideLd0, WideLd1, Acc };631    Function *SMLAD = nullptr;632    if (Exchange)633      SMLAD =634          Acc->getType()->isIntegerTy(32)635              ? Intrinsic::getOrInsertDeclaration(M, Intrinsic::arm_smladx)636              : Intrinsic::getOrInsertDeclaration(M, Intrinsic::arm_smlaldx);637    else638      SMLAD = Acc->getType()->isIntegerTy(32)639                  ? Intrinsic::getOrInsertDeclaration(M, Intrinsic::arm_smlad)640                  : Intrinsic::getOrInsertDeclaration(M, Intrinsic::arm_smlald);641 642    IRBuilder<NoFolder> Builder(InsertAfter->getParent(),643                                BasicBlock::iterator(InsertAfter));644    Instruction *Call = Builder.CreateCall(SMLAD, Args);645    NumSMLAD++;646    return Call;647  };648 649  // Return the instruction after the dominated instruction.650  auto GetInsertPoint = [this](Value *A, Value *B) {651    assert((isa<Instruction>(A) || isa<Instruction>(B)) &&652           "expected at least one instruction");653 654    Value *V = nullptr;655    if (!isa<Instruction>(A))656      V = B;657    else if (!isa<Instruction>(B))658      V = A;659    else660      V = DT->dominates(cast<Instruction>(A), cast<Instruction>(B)) ? B : A;661 662    return &*++BasicBlock::iterator(cast<Instruction>(V));663  };664 665  Value *Acc = R.getAccumulator();666 667  // For any muls that were discovered but not paired, accumulate their values668  // as before.669  IRBuilder<NoFolder> Builder(R.getRoot()->getParent());670  MulCandList &MulCands = R.getMuls();671  for (auto &MulCand : MulCands) {672    if (MulCand->Paired)673      continue;674 675    Instruction *Mul = cast<Instruction>(MulCand->Root);676    LLVM_DEBUG(dbgs() << "Accumulating unpaired mul: " << *Mul << "\n");677 678    if (R.getType() != Mul->getType()) {679      assert(R.is64Bit() && "expected 64-bit result");680      Builder.SetInsertPoint(&*++BasicBlock::iterator(Mul));681      Mul = cast<Instruction>(Builder.CreateSExt(Mul, R.getRoot()->getType()));682    }683 684    if (!Acc) {685      Acc = Mul;686      continue;687    }688 689    // If Acc is the original incoming value to the reduction, it could be a690    // phi. But the phi will dominate Mul, meaning that Mul will be the691    // insertion point.692    Builder.SetInsertPoint(GetInsertPoint(Mul, Acc));693    Acc = Builder.CreateAdd(Mul, Acc);694  }695 696  if (!Acc) {697    Acc = R.is64Bit() ?698      ConstantInt::get(IntegerType::get(M->getContext(), 64), 0) :699      ConstantInt::get(IntegerType::get(M->getContext(), 32), 0);700  } else if (Acc->getType() != R.getType()) {701    Builder.SetInsertPoint(R.getRoot());702    Acc = Builder.CreateSExt(Acc, R.getType());703  }704 705  // Roughly sort the mul pairs in their program order.706  llvm::sort(R.getMulPairs(), [](auto &PairA, auto &PairB) {707    const Instruction *A = PairA.first->Root;708    const Instruction *B = PairB.first->Root;709    return A->comesBefore(B);710  });711 712  IntegerType *Ty = IntegerType::get(M->getContext(), 32);713  for (auto &Pair : R.getMulPairs()) {714    MulCandidate *LHSMul = Pair.first;715    MulCandidate *RHSMul = Pair.second;716    LoadInst *BaseLHS = LHSMul->getBaseLoad();717    LoadInst *BaseRHS = RHSMul->getBaseLoad();718    auto LIt = WideLoads.find(BaseLHS);719    LoadInst *WideLHS = LIt != WideLoads.end()720                            ? LIt->second->getLoad()721                            : CreateWideLoad(LHSMul->VecLd, Ty);722    auto RIt = WideLoads.find(BaseRHS);723    LoadInst *WideRHS = RIt != WideLoads.end()724                            ? RIt->second->getLoad()725                            : CreateWideLoad(RHSMul->VecLd, Ty);726 727    Instruction *InsertAfter = GetInsertPoint(WideLHS, WideRHS);728    InsertAfter = GetInsertPoint(InsertAfter, Acc);729    Acc = CreateSMLAD(WideLHS, WideRHS, Acc, RHSMul->Exchange, InsertAfter);730  }731  R.UpdateRoot(cast<Instruction>(Acc));732}733 734LoadInst* ARMParallelDSP::CreateWideLoad(MemInstList &Loads,735                                         IntegerType *LoadTy) {736  assert(Loads.size() == 2 && "currently only support widening two loads");737 738  LoadInst *Base = Loads[0];739  LoadInst *Offset = Loads[1];740 741  Instruction *BaseSExt = dyn_cast<SExtInst>(Base->user_back());742  Instruction *OffsetSExt = dyn_cast<SExtInst>(Offset->user_back());743 744  assert((BaseSExt && OffsetSExt)745         && "Loads should have a single, extending, user");746 747  std::function<void(Value*, Value*)> MoveBefore =748    [&](Value *A, Value *B) -> void {749      if (!isa<Instruction>(A) || !isa<Instruction>(B))750        return;751 752      auto *Source = cast<Instruction>(A);753      auto *Sink = cast<Instruction>(B);754 755      if (DT->dominates(Source, Sink) ||756          Source->getParent() != Sink->getParent() ||757          isa<PHINode>(Source) || isa<PHINode>(Sink))758        return;759 760      Source->moveBefore(Sink->getIterator());761      for (auto &Op : Source->operands())762        MoveBefore(Op, Source);763    };764 765  // Insert the load at the point of the original dominating load.766  LoadInst *DomLoad = DT->dominates(Base, Offset) ? Base : Offset;767  IRBuilder<NoFolder> IRB(DomLoad->getParent(),768                          ++BasicBlock::iterator(DomLoad));769 770  // Create the wide load, while making sure to maintain the original alignment771  // as this prevents ldrd from being generated when it could be illegal due to772  // memory alignment.773  Value *VecPtr = Base->getPointerOperand();774  LoadInst *WideLoad = IRB.CreateAlignedLoad(LoadTy, VecPtr, Base->getAlign());775 776  // Make sure everything is in the correct order in the basic block.777  MoveBefore(Base->getPointerOperand(), VecPtr);778  MoveBefore(VecPtr, WideLoad);779 780  // From the wide load, create two values that equal the original two loads.781  // Loads[0] needs trunc while Loads[1] needs a lshr and trunc.782  // TODO: Support big-endian as well.783  Value *Bottom = IRB.CreateTrunc(WideLoad, Base->getType());784  Value *NewBaseSExt = IRB.CreateSExt(Bottom, BaseSExt->getType());785  BaseSExt->replaceAllUsesWith(NewBaseSExt);786 787  IntegerType *OffsetTy = cast<IntegerType>(Offset->getType());788  Value *ShiftVal = ConstantInt::get(LoadTy, OffsetTy->getBitWidth());789  Value *Top = IRB.CreateLShr(WideLoad, ShiftVal);790  Value *Trunc = IRB.CreateTrunc(Top, OffsetTy);791  Value *NewOffsetSExt = IRB.CreateSExt(Trunc, OffsetSExt->getType());792  OffsetSExt->replaceAllUsesWith(NewOffsetSExt);793 794  LLVM_DEBUG(dbgs() << "From Base and Offset:\n"795             << *Base << "\n" << *Offset << "\n"796             << "Created Wide Load:\n"797             << *WideLoad << "\n"798             << *Bottom << "\n"799             << *NewBaseSExt << "\n"800             << *Top << "\n"801             << *Trunc << "\n"802             << *NewOffsetSExt << "\n");803  WideLoads.emplace(std::make_pair(Base,804                                   std::make_unique<WidenedLoad>(Loads, WideLoad)));805  return WideLoad;806}807 808Pass *llvm::createARMParallelDSPPass() {809  return new ARMParallelDSP();810}811 812char ARMParallelDSP::ID = 0;813 814INITIALIZE_PASS_BEGIN(ARMParallelDSP, "arm-parallel-dsp",815                "Transform functions to use DSP intrinsics", false, false)816INITIALIZE_PASS_END(ARMParallelDSP, "arm-parallel-dsp",817                "Transform functions to use DSP intrinsics", false, false)818