lib/Transforms/Scalar/LoopUnrollAndJamPass.cpp

   1 //===- LoopUnrollAndJam.cpp - Loop unroll and jam 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-exception
   6 //
   7 //===----------------------------------------------------------------------===//
   8 //
   9 // This pass implements an unroll and jam pass. Most of the work is done by
  10 // Utils/UnrollLoopAndJam.cpp.
  11 //===----------------------------------------------------------------------===//
  12
  13 #include "llvm/Transforms/Scalar/LoopUnrollAndJamPass.h"
  14 #include "llvm/ADT/None.h"
  15 #include "llvm/ADT/STLExtras.h"
  16 #include "llvm/ADT/SmallPtrSet.h"
  17 #include "llvm/ADT/StringRef.h"
  18 #include "llvm/Analysis/AssumptionCache.h"
  19 #include "llvm/Analysis/CodeMetrics.h"
  20 #include "llvm/Analysis/DependenceAnalysis.h"
  21 #include "llvm/Analysis/LoopAnalysisManager.h"
  22 #include "llvm/Analysis/LoopInfo.h"
  23 #include "llvm/Analysis/LoopPass.h"
  24 #include "llvm/Analysis/OptimizationRemarkEmitter.h"
  25 #include "llvm/Analysis/ScalarEvolution.h"
  26 #include "llvm/Analysis/TargetTransformInfo.h"
  27 #include "llvm/IR/BasicBlock.h"
  28 #include "llvm/IR/CFG.h"
  29 #include "llvm/IR/Constant.h"
  30 #include "llvm/IR/Constants.h"
  31 #include "llvm/IR/Dominators.h"
  32 #include "llvm/IR/Function.h"
  33 #include "llvm/IR/Instruction.h"
  34 #include "llvm/IR/Instructions.h"
  35 #include "llvm/IR/IntrinsicInst.h"
  36 #include "llvm/IR/Metadata.h"
  37 #include "llvm/IR/PassManager.h"
  38 #include "llvm/Pass.h"
  39 #include "llvm/Support/Casting.h"
  40 #include "llvm/Support/CommandLine.h"
  41 #include "llvm/Support/Debug.h"
  42 #include "llvm/Support/ErrorHandling.h"
  43 #include "llvm/Support/raw_ostream.h"
  44 #include "llvm/Transforms/Scalar.h"
  45 #include "llvm/Transforms/Scalar/LoopPassManager.h"
  46 #include "llvm/Transforms/Utils.h"
  47 #include "llvm/Transforms/Utils/LoopUtils.h"
  48 #include "llvm/Transforms/Utils/UnrollLoop.h"
  49 #include <algorithm>
  50 #include <cassert>
  51 #include <cstdint>
  52 #include <string>
  53
  54 using namespace llvm;
  55
  56 #define DEBUG_TYPE "loop-unroll-and-jam"
  57
  58 /// @{
  59 /// Metadata attribute names
  60 static const char *const LLVMLoopUnrollAndJamFollowupAll =
  61     "llvm.loop.unroll_and_jam.followup_all";
  62 static const char *const LLVMLoopUnrollAndJamFollowupInner =
  63     "llvm.loop.unroll_and_jam.followup_inner";
  64 static const char *const LLVMLoopUnrollAndJamFollowupOuter =
  65     "llvm.loop.unroll_and_jam.followup_outer";
  66 static const char *const LLVMLoopUnrollAndJamFollowupRemainderInner =
  67     "llvm.loop.unroll_and_jam.followup_remainder_inner";
  68 static const char *const LLVMLoopUnrollAndJamFollowupRemainderOuter =
  69     "llvm.loop.unroll_and_jam.followup_remainder_outer";
  70 /// @}
  71
  72 static cl::opt<bool>
  73     AllowUnrollAndJam("allow-unroll-and-jam", cl::Hidden,
  74                       cl::desc("Allows loops to be unroll-and-jammed."));
  75
  76 static cl::opt<unsigned> UnrollAndJamCount(
  77     "unroll-and-jam-count", cl::Hidden,
  78     cl::desc("Use this unroll count for all loops including those with "
  79              "unroll_and_jam_count pragma values, for testing purposes"));
  80
  81 static cl::opt<unsigned> UnrollAndJamThreshold(
  82     "unroll-and-jam-threshold", cl::init(60), cl::Hidden,
  83     cl::desc("Threshold to use for inner loop when doing unroll and jam."));
  84
  85 static cl::opt<unsigned> PragmaUnrollAndJamThreshold(
  86     "pragma-unroll-and-jam-threshold", cl::init(1024), cl::Hidden,
  87     cl::desc("Unrolled size limit for loops with an unroll_and_jam(full) or "
  88              "unroll_count pragma."));
  89
  90 // Returns the loop hint metadata node with the given name (for example,
  91 // "llvm.loop.unroll.count").  If no such metadata node exists, then nullptr is
  92 // returned.
  93 static MDNode *GetUnrollMetadataForLoop(const Loop *L, StringRef Name) {
  94   if (MDNode *LoopID = L->getLoopID())
  95     return GetUnrollMetadata(LoopID, Name);
  96   return nullptr;
  97 }
  98
  99 // Returns true if the loop has any metadata starting with Prefix. For example a
 100 // Prefix of "llvm.loop.unroll." returns true if we have any unroll metadata.
 101 static bool HasAnyUnrollPragma(const Loop *L, StringRef Prefix) {
 102   if (MDNode *LoopID = L->getLoopID()) {
 103     // First operand should refer to the loop id itself.
 104     assert(LoopID->getNumOperands() > 0 && "requires at least one operand");
 105     assert(LoopID->getOperand(0) == LoopID && "invalid loop id");
 106
 107     for (unsigned i = 1, e = LoopID->getNumOperands(); i < e; ++i) {
 108       MDNode *MD = dyn_cast<MDNode>(LoopID->getOperand(i));
 109       if (!MD)
 110         continue;
 111
 112       MDString *S = dyn_cast<MDString>(MD->getOperand(0));
 113       if (!S)
 114         continue;
 115
 116       if (S->getString().startswith(Prefix))
 117         return true;
 118     }
 119   }
 120   return false;
 121 }
 122
 123 // Returns true if the loop has an unroll_and_jam(enable) pragma.
 124 static bool HasUnrollAndJamEnablePragma(const Loop *L) {
 125   return GetUnrollMetadataForLoop(L, "llvm.loop.unroll_and_jam.enable");
 126 }
 127
 128 // If loop has an unroll_and_jam_count pragma return the (necessarily
 129 // positive) value from the pragma.  Otherwise return 0.
 130 static unsigned UnrollAndJamCountPragmaValue(const Loop *L) {
 131   MDNode *MD = GetUnrollMetadataForLoop(L, "llvm.loop.unroll_and_jam.count");
 132   if (MD) {
 133     assert(MD->getNumOperands() == 2 &&
 134            "Unroll count hint metadata should have two operands.");
 135     unsigned Count =
 136         mdconst::extract<ConstantInt>(MD->getOperand(1))->getZExtValue();
 137     assert(Count >= 1 && "Unroll count must be positive.");
 138     return Count;
 139   }
 140   return 0;
 141 }
 142
 143 // Returns loop size estimation for unrolled loop.
 144 static uint64_t
 145 getUnrollAndJammedLoopSize(unsigned LoopSize,
 146                            TargetTransformInfo::UnrollingPreferences &UP) {
 147   assert(LoopSize >= UP.BEInsns && "LoopSize should not be less than BEInsns!");
 148   return static_cast<uint64_t>(LoopSize - UP.BEInsns) * UP.Count + UP.BEInsns;
 149 }
 150
 151 // Calculates unroll and jam count and writes it to UP.Count. Returns true if
 152 // unroll count was set explicitly.
 153 static bool computeUnrollAndJamCount(
 154     Loop *L, Loop *SubLoop, const TargetTransformInfo &TTI, DominatorTree &DT,
 155     LoopInfo *LI, ScalarEvolution &SE,
 156     const SmallPtrSetImpl<const Value *> &EphValues,
 157     OptimizationRemarkEmitter *ORE, unsigned OuterTripCount,
 158     unsigned OuterTripMultiple, unsigned OuterLoopSize, unsigned InnerTripCount,
 159     unsigned InnerLoopSize, TargetTransformInfo::UnrollingPreferences &UP) {
 160   // First up use computeUnrollCount from the loop unroller to get a count
 161   // for unrolling the outer loop, plus any loops requiring explicit
 162   // unrolling we leave to the unroller. This uses UP.Threshold /
 163   // UP.PartialThreshold / UP.MaxCount to come up with sensible loop values.
 164   // We have already checked that the loop has no unroll.* pragmas.
 165   unsigned MaxTripCount = 0;
 166   bool UseUpperBound = false;
 167   bool ExplicitUnroll = computeUnrollCount(
 168       L, TTI, DT, LI, SE, EphValues, ORE, OuterTripCount, MaxTripCount,
 169       OuterTripMultiple, OuterLoopSize, UP, UseUpperBound);
 170   if (ExplicitUnroll || UseUpperBound) {
 171     // If the user explicitly set the loop as unrolled, dont UnJ it. Leave it
 172     // for the unroller instead.
 173     LLVM_DEBUG(dbgs() << "Won't unroll-and-jam; explicit count set by "
 174                          "computeUnrollCount\n");
 175     UP.Count = 0;
 176     return false;
 177   }
 178
 179   // Override with any explicit Count from the "unroll-and-jam-count" option.
 180   bool UserUnrollCount = UnrollAndJamCount.getNumOccurrences() > 0;
 181   if (UserUnrollCount) {
 182     UP.Count = UnrollAndJamCount;
 183     UP.Force = true;
 184     if (UP.AllowRemainder &&
 185         getUnrollAndJammedLoopSize(OuterLoopSize, UP) < UP.Threshold &&
 186         getUnrollAndJammedLoopSize(InnerLoopSize, UP) <
 187             UP.UnrollAndJamInnerLoopThreshold)
 188       return true;
 189   }
 190
 191   // Check for unroll_and_jam pragmas
 192   unsigned PragmaCount = UnrollAndJamCountPragmaValue(L);
 193   if (PragmaCount > 0) {
 194     UP.Count = PragmaCount;
 195     UP.Runtime = true;
 196     UP.Force = true;
 197     if ((UP.AllowRemainder || (OuterTripMultiple % PragmaCount == 0)) &&
 198         getUnrollAndJammedLoopSize(OuterLoopSize, UP) < UP.Threshold &&
 199         getUnrollAndJammedLoopSize(InnerLoopSize, UP) <
 200             UP.UnrollAndJamInnerLoopThreshold)
 201       return true;
 202   }
 203
 204   bool PragmaEnableUnroll = HasUnrollAndJamEnablePragma(L);
 205   bool ExplicitUnrollAndJamCount = PragmaCount > 0 || UserUnrollCount;
 206   bool ExplicitUnrollAndJam = PragmaEnableUnroll || ExplicitUnrollAndJamCount;
 207
 208   // If the loop has an unrolling pragma, we want to be more aggressive with
 209   // unrolling limits.
 210   if (ExplicitUnrollAndJam)
 211     UP.UnrollAndJamInnerLoopThreshold = PragmaUnrollAndJamThreshold;
 212
 213   if (!UP.AllowRemainder && getUnrollAndJammedLoopSize(InnerLoopSize, UP) >=
 214                                 UP.UnrollAndJamInnerLoopThreshold) {
 215     LLVM_DEBUG(dbgs() << "Won't unroll-and-jam; can't create remainder and "
 216                          "inner loop too large\n");
 217     UP.Count = 0;
 218     return false;
 219   }
 220
 221   // We have a sensible limit for the outer loop, now adjust it for the inner
 222   // loop and UP.UnrollAndJamInnerLoopThreshold. If the outer limit was set
 223   // explicitly, we want to stick to it.
 224   if (!ExplicitUnrollAndJamCount && UP.AllowRemainder) {
 225     while (UP.Count != 0 && getUnrollAndJammedLoopSize(InnerLoopSize, UP) >=
 226                                 UP.UnrollAndJamInnerLoopThreshold)
 227       UP.Count--;
 228   }
 229
 230   // If we are explicitly unroll and jamming, we are done. Otherwise there are a
 231   // number of extra performance heuristics to check.
 232   if (ExplicitUnrollAndJam)
 233     return true;
 234
 235   // If the inner loop count is known and small, leave the entire loop nest to
 236   // be the unroller
 237   if (InnerTripCount && InnerLoopSize * InnerTripCount < UP.Threshold) {
 238     LLVM_DEBUG(dbgs() << "Won't unroll-and-jam; small inner loop count is "
 239                          "being left for the unroller\n");
 240     UP.Count = 0;
 241     return false;
 242   }
 243
 244   // Check for situations where UnJ is likely to be unprofitable. Including
 245   // subloops with more than 1 block.
 246   if (SubLoop->getBlocks().size() != 1) {
 247     LLVM_DEBUG(
 248         dbgs() << "Won't unroll-and-jam; More than one inner loop block\n");
 249     UP.Count = 0;
 250     return false;
 251   }
 252
 253   // Limit to loops where there is something to gain from unrolling and
 254   // jamming the loop. In this case, look for loads that are invariant in the
 255   // outer loop and can become shared.
 256   unsigned NumInvariant = 0;
 257   for (BasicBlock *BB : SubLoop->getBlocks()) {
 258     for (Instruction &I : *BB) {
 259       if (auto *Ld = dyn_cast<LoadInst>(&I)) {
 260         Value *V = Ld->getPointerOperand();
 261         const SCEV *LSCEV = SE.getSCEVAtScope(V, L);
 262         if (SE.isLoopInvariant(LSCEV, L))
 263           NumInvariant++;
 264       }
 265     }
 266   }
 267   if (NumInvariant == 0) {
 268     LLVM_DEBUG(dbgs() << "Won't unroll-and-jam; No loop invariant loads\n");
 269     UP.Count = 0;
 270     return false;
 271   }
 272
 273   return false;
 274 }
 275
 276 static LoopUnrollResult
 277 tryToUnrollAndJamLoop(Loop *L, DominatorTree &DT, LoopInfo *LI,
 278                       ScalarEvolution &SE, const TargetTransformInfo &TTI,
 279                       AssumptionCache &AC, DependenceInfo &DI,
 280                       OptimizationRemarkEmitter &ORE, int OptLevel) {
 281   // Quick checks of the correct loop form
 282   if (!L->isLoopSimplifyForm() || L->getSubLoops().size() != 1)
 283     return LoopUnrollResult::Unmodified;
 284   Loop *SubLoop = L->getSubLoops()[0];
 285   if (!SubLoop->isLoopSimplifyForm())
 286     return LoopUnrollResult::Unmodified;
 287
 288   BasicBlock *Latch = L->getLoopLatch();
 289   BasicBlock *Exit = L->getExitingBlock();
 290   BasicBlock *SubLoopLatch = SubLoop->getLoopLatch();
 291   BasicBlock *SubLoopExit = SubLoop->getExitingBlock();
 292
 293   if (Latch != Exit || SubLoopLatch != SubLoopExit)
 294     return LoopUnrollResult::Unmodified;
 295
 296   TargetTransformInfo::UnrollingPreferences UP = gatherUnrollingPreferences(
 297       L, SE, TTI, OptLevel, None, None, None, None, None, None);
 298   if (AllowUnrollAndJam.getNumOccurrences() > 0)
 299     UP.UnrollAndJam = AllowUnrollAndJam;
 300   if (UnrollAndJamThreshold.getNumOccurrences() > 0)
 301     UP.UnrollAndJamInnerLoopThreshold = UnrollAndJamThreshold;
 302   // Exit early if unrolling is disabled.
 303   if (!UP.UnrollAndJam || UP.UnrollAndJamInnerLoopThreshold == 0)
 304     return LoopUnrollResult::Unmodified;
 305
 306   LLVM_DEBUG(dbgs() << "Loop Unroll and Jam: F["
 307                     << L->getHeader()->getParent()->getName() << "] Loop %"
 308                     << L->getHeader()->getName() << "\n");
 309
 310   TransformationMode EnableMode = hasUnrollAndJamTransformation(L);
 311   if (EnableMode & TM_Disable)
 312     return LoopUnrollResult::Unmodified;
 313
 314   // A loop with any unroll pragma (enabling/disabling/count/etc) is left for
 315   // the unroller, so long as it does not explicitly have unroll_and_jam
 316   // metadata. This means #pragma nounroll will disable unroll and jam as well
 317   // as unrolling
 318   if (HasAnyUnrollPragma(L, "llvm.loop.unroll.") &&
 319       !HasAnyUnrollPragma(L, "llvm.loop.unroll_and_jam.")) {
 320     LLVM_DEBUG(dbgs() << "  Disabled due to pragma.\n");
 321     return LoopUnrollResult::Unmodified;
 322   }
 323
 324   if (!isSafeToUnrollAndJam(L, SE, DT, DI)) {
 325     LLVM_DEBUG(dbgs() << "  Disabled due to not being safe.\n");
 326     return LoopUnrollResult::Unmodified;
 327   }
 328
 329   // Approximate the loop size and collect useful info
 330   unsigned NumInlineCandidates;
 331   bool NotDuplicatable;
 332   bool Convergent;
 333   SmallPtrSet<const Value *, 32> EphValues;
 334   CodeMetrics::collectEphemeralValues(L, &AC, EphValues);
 335   unsigned InnerLoopSize =
 336       ApproximateLoopSize(SubLoop, NumInlineCandidates, NotDuplicatable,
 337                           Convergent, TTI, EphValues, UP.BEInsns);
 338   unsigned OuterLoopSize =
 339       ApproximateLoopSize(L, NumInlineCandidates, NotDuplicatable, Convergent,
 340                           TTI, EphValues, UP.BEInsns);
 341   LLVM_DEBUG(dbgs() << "  Outer Loop Size: " << OuterLoopSize << "\n");
 342   LLVM_DEBUG(dbgs() << "  Inner Loop Size: " << InnerLoopSize << "\n");
 343   if (NotDuplicatable) {
 344     LLVM_DEBUG(dbgs() << "  Not unrolling loop which contains non-duplicatable "
 345                          "instructions.\n");
 346     return LoopUnrollResult::Unmodified;
 347   }
 348   if (NumInlineCandidates != 0) {
 349     LLVM_DEBUG(dbgs() << "  Not unrolling loop with inlinable calls.\n");
 350     return LoopUnrollResult::Unmodified;
 351   }
 352   if (Convergent) {
 353     LLVM_DEBUG(
 354         dbgs() << "  Not unrolling loop with convergent instructions.\n");
 355     return LoopUnrollResult::Unmodified;
 356   }
 357
 358   // Save original loop IDs for after the transformation.
 359   MDNode *OrigOuterLoopID = L->getLoopID();
 360   MDNode *OrigSubLoopID = SubLoop->getLoopID();
 361
 362   // To assign the loop id of the epilogue, assign it before unrolling it so it
 363   // is applied to every inner loop of the epilogue. We later apply the loop ID
 364   // for the jammed inner loop.
 365   Optional<MDNode *> NewInnerEpilogueLoopID = makeFollowupLoopID(
 366       OrigOuterLoopID, {LLVMLoopUnrollAndJamFollowupAll,
 367                         LLVMLoopUnrollAndJamFollowupRemainderInner});
 368   if (NewInnerEpilogueLoopID.hasValue())
 369     SubLoop->setLoopID(NewInnerEpilogueLoopID.getValue());
 370
 371   // Find trip count and trip multiple
 372   unsigned OuterTripCount = SE.getSmallConstantTripCount(L, Latch);
 373   unsigned OuterTripMultiple = SE.getSmallConstantTripMultiple(L, Latch);
 374   unsigned InnerTripCount = SE.getSmallConstantTripCount(SubLoop, SubLoopLatch);
 375
 376   // Decide if, and by how much, to unroll
 377   bool IsCountSetExplicitly = computeUnrollAndJamCount(
 378       L, SubLoop, TTI, DT, LI, SE, EphValues, &ORE, OuterTripCount,
 379       OuterTripMultiple, OuterLoopSize, InnerTripCount, InnerLoopSize, UP);
 380   if (UP.Count <= 1)
 381     return LoopUnrollResult::Unmodified;
 382   // Unroll factor (Count) must be less or equal to TripCount.
 383   if (OuterTripCount && UP.Count > OuterTripCount)
 384     UP.Count = OuterTripCount;
 385
 386   Loop *EpilogueOuterLoop = nullptr;
 387   LoopUnrollResult UnrollResult = UnrollAndJamLoop(
 388       L, UP.Count, OuterTripCount, OuterTripMultiple, UP.UnrollRemainder, LI,
 389       &SE, &DT, &AC, &ORE, &EpilogueOuterLoop);
 390
 391   // Assign new loop attributes.
 392   if (EpilogueOuterLoop) {
 393     Optional<MDNode *> NewOuterEpilogueLoopID = makeFollowupLoopID(
 394         OrigOuterLoopID, {LLVMLoopUnrollAndJamFollowupAll,
 395                           LLVMLoopUnrollAndJamFollowupRemainderOuter});
 396     if (NewOuterEpilogueLoopID.hasValue())
 397       EpilogueOuterLoop->setLoopID(NewOuterEpilogueLoopID.getValue());
 398   }
 399
 400   Optional<MDNode *> NewInnerLoopID =
 401       makeFollowupLoopID(OrigOuterLoopID, {LLVMLoopUnrollAndJamFollowupAll,
 402                                            LLVMLoopUnrollAndJamFollowupInner});
 403   if (NewInnerLoopID.hasValue())
 404     SubLoop->setLoopID(NewInnerLoopID.getValue());
 405   else
 406     SubLoop->setLoopID(OrigSubLoopID);
 407
 408   if (UnrollResult == LoopUnrollResult::PartiallyUnrolled) {
 409     Optional<MDNode *> NewOuterLoopID = makeFollowupLoopID(
 410         OrigOuterLoopID,
 411         {LLVMLoopUnrollAndJamFollowupAll, LLVMLoopUnrollAndJamFollowupOuter});
 412     if (NewOuterLoopID.hasValue()) {
 413       L->setLoopID(NewOuterLoopID.getValue());
 414
 415       // Do not setLoopAlreadyUnrolled if a followup was given.
 416       return UnrollResult;
 417     }
 418   }
 419
 420   // If loop has an unroll count pragma or unrolled by explicitly set count
 421   // mark loop as unrolled to prevent unrolling beyond that requested.
 422   if (UnrollResult != LoopUnrollResult::FullyUnrolled && IsCountSetExplicitly)
 423     L->setLoopAlreadyUnrolled();
 424
 425   return UnrollResult;
 426 }
 427
 428 namespace {
 429
 430 class LoopUnrollAndJam : public LoopPass {
 431 public:
 432   static char ID; // Pass ID, replacement for typeid
 433   unsigned OptLevel;
 434
 435   LoopUnrollAndJam(int OptLevel = 2) : LoopPass(ID), OptLevel(OptLevel) {
 436     initializeLoopUnrollAndJamPass(*PassRegistry::getPassRegistry());
 437   }
 438
 439   bool runOnLoop(Loop *L, LPPassManager &LPM) override {
 440     if (skipLoop(L))
 441       return false;
 442
 443     Function &F = *L->getHeader()->getParent();
 444
 445     auto &DT = getAnalysis<DominatorTreeWrapperPass>().getDomTree();
 446     LoopInfo *LI = &getAnalysis<LoopInfoWrapperPass>().getLoopInfo();
 447     ScalarEvolution &SE = getAnalysis<ScalarEvolutionWrapperPass>().getSE();
 448     const TargetTransformInfo &TTI =
 449         getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F);
 450     auto &AC = getAnalysis<AssumptionCacheTracker>().getAssumptionCache(F);
 451     auto &DI = getAnalysis<DependenceAnalysisWrapperPass>().getDI();
 452     // For the old PM, we can't use OptimizationRemarkEmitter as an analysis
 453     // pass.  Function analyses need to be preserved across loop transformations
 454     // but ORE cannot be preserved (see comment before the pass definition).
 455     OptimizationRemarkEmitter ORE(&F);
 456
 457     LoopUnrollResult Result =
 458         tryToUnrollAndJamLoop(L, DT, LI, SE, TTI, AC, DI, ORE, OptLevel);
 459
 460     if (Result == LoopUnrollResult::FullyUnrolled)
 461       LPM.markLoopAsDeleted(*L);
 462
 463     return Result != LoopUnrollResult::Unmodified;
 464   }
 465
 466   /// This transformation requires natural loop information & requires that
 467   /// loop preheaders be inserted into the CFG...
 468   void getAnalysisUsage(AnalysisUsage &AU) const override {
 469     AU.addRequired<AssumptionCacheTracker>();
 470     AU.addRequired<TargetTransformInfoWrapperPass>();
 471     AU.addRequired<DependenceAnalysisWrapperPass>();
 472     getLoopAnalysisUsage(AU);
 473   }
 474 };
 475
 476 } // end anonymous namespace
 477
 478 char LoopUnrollAndJam::ID = 0;
 479
 480 INITIALIZE_PASS_BEGIN(LoopUnrollAndJam, "loop-unroll-and-jam",
 481                       "Unroll and Jam loops", false, false)
 482 INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)
 483 INITIALIZE_PASS_DEPENDENCY(LoopPass)
 484 INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass)
 485 INITIALIZE_PASS_DEPENDENCY(DependenceAnalysisWrapperPass)
 486 INITIALIZE_PASS_END(LoopUnrollAndJam, "loop-unroll-and-jam",
 487                     "Unroll and Jam loops", false, false)
 488
 489 Pass *llvm::createLoopUnrollAndJamPass(int OptLevel) {
 490   return new LoopUnrollAndJam(OptLevel);
 491 }
 492
 493 PreservedAnalyses LoopUnrollAndJamPass::run(Loop &L, LoopAnalysisManager &AM,
 494                                             LoopStandardAnalysisResults &AR,
 495                                             LPMUpdater &) {
 496   const auto &FAM =
 497       AM.getResult<FunctionAnalysisManagerLoopProxy>(L, AR).getManager();
 498   Function *F = L.getHeader()->getParent();
 499
 500   auto *ORE = FAM.getCachedResult<OptimizationRemarkEmitterAnalysis>(*F);
 501   // FIXME: This should probably be optional rather than required.
 502   if (!ORE)
 503     report_fatal_error(
 504         "LoopUnrollAndJamPass: OptimizationRemarkEmitterAnalysis not cached at "
 505         "a higher level");
 506
 507   DependenceInfo DI(F, &AR.AA, &AR.SE, &AR.LI);
 508
 509   LoopUnrollResult Result = tryToUnrollAndJamLoop(
 510       &L, AR.DT, &AR.LI, AR.SE, AR.TTI, AR.AC, DI, *ORE, OptLevel);
 511
 512   if (Result == LoopUnrollResult::Unmodified)
 513     return PreservedAnalyses::all();
 514
 515   return getLoopPassPreservedAnalyses();
 516 }