lib/Transforms/Scalar/LoopUnroll.cpp

   1 //===-- LoopUnroll.cpp - Loop unroller pass -------------------------------===//
   2 //
   3 //                     The LLVM Compiler Infrastructure
   4 //
   5 // This file is distributed under the University of Illinois Open Source
   6 // License. See LICENSE.TXT for details.
   7 //
   8 //===----------------------------------------------------------------------===//
   9 //
  10 // This pass implements a simple loop unroller.  It works best when loops have
  11 // been canonicalized by the -indvars pass, allowing it to determine the trip
  12 // counts of loops easily.
  13 //===----------------------------------------------------------------------===//
  14
  15 #define DEBUG_TYPE "loop-unroll"
  16 #include "llvm/IntrinsicInst.h"
  17 #include "llvm/Transforms/Scalar.h"
  18 #include "llvm/Analysis/LoopInfo.h"
  19 #include "llvm/Analysis/LoopPass.h"
  20 #include "llvm/Support/CommandLine.h"
  21 #include "llvm/Support/Debug.h"
  22 #include "llvm/Support/raw_ostream.h"
  23 #include "llvm/Transforms/Utils/UnrollLoop.h"
  24 #include <climits>
  25
  26 using namespace llvm;
  27
  28 static cl::opt<unsigned>
  29 UnrollThreshold("unroll-threshold", cl::init(100), cl::Hidden,
  30   cl::desc("The cut-off point for automatic loop unrolling"));
  31
  32 static cl::opt<unsigned>
  33 UnrollCount("unroll-count", cl::init(0), cl::Hidden,
  34   cl::desc("Use this unroll count for all loops, for testing purposes"));
  35
  36 static cl::opt<bool>
  37 UnrollAllowPartial("unroll-allow-partial", cl::init(false), cl::Hidden,
  38   cl::desc("Allows loops to be partially unrolled until "
  39            "-unroll-threshold loop size is reached."));
  40
  41 namespace {
  42   class LoopUnroll : public LoopPass {
  43   public:
  44     static char ID; // Pass ID, replacement for typeid
  45     LoopUnroll() : LoopPass(&ID) {}
  46
  47     /// A magic value for use with the Threshold parameter to indicate
  48     /// that the loop unroll should be performed regardless of how much
  49     /// code expansion would result.
  50     static const unsigned NoThreshold = UINT_MAX;
  51
  52     bool runOnLoop(Loop *L, LPPassManager &LPM);
  53
  54     /// This transformation requires natural loop information & requires that
  55     /// loop preheaders be inserted into the CFG...
  56     ///
  57     virtual void getAnalysisUsage(AnalysisUsage &AU) const {
  58       AU.addRequiredID(LoopSimplifyID);
  59       AU.addRequiredID(LCSSAID);
  60       AU.addRequired<LoopInfo>();
  61       AU.addPreservedID(LCSSAID);
  62       AU.addPreserved<LoopInfo>();
  63       // FIXME: Loop unroll requires LCSSA. And LCSSA requires dom info.
  64       // If loop unroll does not preserve dom info then LCSSA pass on next
  65       // loop will receive invalid dom info.
  66       // For now, recreate dom info, if loop is unrolled.
  67       AU.addPreserved<DominatorTree>();
  68       AU.addPreserved<DominanceFrontier>();
  69     }
  70   };
  71 }
  72
  73 char LoopUnroll::ID = 0;
  74 static RegisterPass<LoopUnroll> X("loop-unroll", "Unroll loops");
  75
  76 Pass *llvm::createLoopUnrollPass() { return new LoopUnroll(); }
  77
  78 /// ApproximateLoopSize - Approximate the size of the loop.
  79 static unsigned ApproximateLoopSize(const Loop *L) {
  80   unsigned Size = 0;
  81   for (Loop::block_iterator I = L->block_begin(), E = L->block_end();
  82        I != E; ++I) {
  83     BasicBlock *BB = *I;
  84     Instruction *Term = BB->getTerminator();
  85     for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
  86       if (isa<PHINode>(I) && BB == L->getHeader()) {
  87         // Ignore PHI nodes in the header.
  88       } else if (I->hasOneUse() && I->use_back() == Term) {
  89         // Ignore instructions only used by the loop terminator.
  90       } else if (isa<DbgInfoIntrinsic>(I)) {
  91         // Ignore debug instructions
  92       } else if (isa<GetElementPtrInst>(I) && I->hasOneUse()) {
  93         // Ignore GEP as they generally are subsumed into a load or store.
  94       } else if (isa<CallInst>(I)) {
  95         // Estimate size overhead introduced by call instructions which
  96         // is higher than other instructions. Here 3 and 10 are magic
  97         // numbers that help one isolated test case from PR2067 without
  98         // negatively impacting measured benchmarks.
  99         Size += isa<IntrinsicInst>(I) ? 3 : 10;
 100       } else {
 101         ++Size;
 102       }
 103
 104       // TODO: Ignore expressions derived from PHI and constants if inval of phi
 105       // is a constant, or if operation is associative.  This will get induction
 106       // variables.
 107     }
 108   }
 109
 110   return Size;
 111 }
 112
 113 bool LoopUnroll::runOnLoop(Loop *L, LPPassManager &LPM) {
 114   assert(L->isLCSSAForm());
 115   LoopInfo *LI = &getAnalysis<LoopInfo>();
 116
 117   BasicBlock *Header = L->getHeader();
 118   DEBUG(errs() << "Loop Unroll: F[" << Header->getParent()->getName()
 119         << "] Loop %" << Header->getName() << "\n");
 120   (void)Header;
 121
 122   // Find trip count
 123   unsigned TripCount = L->getSmallConstantTripCount();
 124   unsigned Count = UnrollCount;
 125
 126   // Automatically select an unroll count.
 127   if (Count == 0) {
 128     // Conservative heuristic: if we know the trip count, see if we can
 129     // completely unroll (subject to the threshold, checked below); otherwise
 130     // try to find greatest modulo of the trip count which is still under
 131     // threshold value.
 132     if (TripCount == 0)
 133       return false;
 134     Count = TripCount;
 135   }
 136
 137   // Enforce the threshold.
 138   if (UnrollThreshold != NoThreshold) {
 139     unsigned LoopSize = ApproximateLoopSize(L);
 140     DEBUG(errs() << "  Loop Size = " << LoopSize << "\n");
 141     uint64_t Size = (uint64_t)LoopSize*Count;
 142     if (TripCount != 1 && Size > UnrollThreshold) {
 143       DEBUG(errs() << "  Too large to fully unroll with count: " << Count
 144             << " because size: " << Size << ">" << UnrollThreshold << "\n");
 145       if (!UnrollAllowPartial) {
 146         DEBUG(errs() << "  will not try to unroll partially because "
 147               << "-unroll-allow-partial not given\n");
 148         return false;
 149       }
 150       // Reduce unroll count to be modulo of TripCount for partial unrolling
 151       Count = UnrollThreshold / LoopSize;
 152       while (Count != 0 && TripCount%Count != 0) {
 153         Count--;
 154       }
 155       if (Count < 2) {
 156         DEBUG(errs() << "  could not unroll partially\n");
 157         return false;
 158       }
 159       DEBUG(errs() << "  partially unrolling with count: " << Count << "\n");
 160     }
 161   }
 162
 163   // Unroll the loop.
 164   Function *F = L->getHeader()->getParent();
 165   if (!UnrollLoop(L, Count, LI, &LPM))
 166     return false;
 167
 168   // FIXME: Reconstruct dom info, because it is not preserved properly.
 169   DominatorTree *DT = getAnalysisIfAvailable<DominatorTree>();
 170   if (DT) {
 171     DT->runOnFunction(*F);
 172     DominanceFrontier *DF = getAnalysisIfAvailable<DominanceFrontier>();
 173     if (DF)
 174       DF->runOnFunction(*F);
 175   }
 176   return true;
 177 }