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/Compiler.h"
  21 #include "llvm/Support/CommandLine.h"
  22 #include "llvm/Support/Debug.h"
  23 #include "llvm/Support/raw_ostream.h"
  24 #include "llvm/Transforms/Utils/UnrollLoop.h"
  25 #include <climits>
  26
  27 using namespace llvm;
  28
  29 static cl::opt<unsigned>
  30 UnrollThreshold("unroll-threshold", cl::init(100), cl::Hidden,
  31   cl::desc("The cut-off point for automatic loop unrolling"));
  32
  33 static cl::opt<unsigned>
  34 UnrollCount("unroll-count", cl::init(0), cl::Hidden,
  35   cl::desc("Use this unroll count for all loops, for testing purposes"));
  36
  37 static cl::opt<bool>
  38 UnrollAllowPartial("unroll-allow-partial", cl::init(false), cl::Hidden,
  39   cl::desc("Allows loops to be partially unrolled until "
  40            "-unroll-threshold loop size is reached."));
  41
  42 namespace {
  43   class VISIBILITY_HIDDEN LoopUnroll : public LoopPass {
  44   public:
  45     static char ID; // Pass ID, replacement for typeid
  46     LoopUnroll() : LoopPass(&ID) {}
  47
  48     /// A magic value for use with the Threshold parameter to indicate
  49     /// that the loop unroll should be performed regardless of how much
  50     /// code expansion would result.
  51     static const unsigned NoThreshold = UINT_MAX;
  52
  53     bool runOnLoop(Loop *L, LPPassManager &LPM);
  54
  55     /// This transformation requires natural loop information & requires that
  56     /// loop preheaders be inserted into the CFG...
  57     ///
  58     virtual void getAnalysisUsage(AnalysisUsage &AU) const {
  59       AU.addRequiredID(LoopSimplifyID);
  60       AU.addRequiredID(LCSSAID);
  61       AU.addRequired<LoopInfo>();
  62       AU.addPreservedID(LCSSAID);
  63       AU.addPreserved<LoopInfo>();
  64       // FIXME: Loop unroll requires LCSSA. And LCSSA requires dom info.
  65       // If loop unroll does not preserve dom info then LCSSA pass on next
  66       // loop will receive invalid dom info.
  67       // For now, recreate dom info, if loop is unrolled.
  68       AU.addPreserved<DominatorTree>();
  69       AU.addPreserved<DominanceFrontier>();
  70     }
  71   };
  72 }
  73
  74 char LoopUnroll::ID = 0;
  75 static RegisterPass<LoopUnroll> X("loop-unroll", "Unroll loops");
  76
  77 Pass *llvm::createLoopUnrollPass() { return new LoopUnroll(); }
  78
  79 /// ApproximateLoopSize - Approximate the size of the loop.
  80 static unsigned ApproximateLoopSize(const Loop *L) {
  81   unsigned Size = 0;
  82   for (Loop::block_iterator I = L->block_begin(), E = L->block_end();
  83        I != E; ++I) {
  84     BasicBlock *BB = *I;
  85     Instruction *Term = BB->getTerminator();
  86     for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
  87       if (isa<PHINode>(I) && BB == L->getHeader()) {
  88         // Ignore PHI nodes in the header.
  89       } else if (I->hasOneUse() && I->use_back() == Term) {
  90         // Ignore instructions only used by the loop terminator.
  91       } else if (isa<DbgInfoIntrinsic>(I)) {
  92         // Ignore debug instructions
  93       } else if (isa<GetElementPtrInst>(I) && I->hasOneUse()) {
  94         // Ignore GEP as they generally are subsumed into a load or store.
  95       } else if (isa<CallInst>(I)) {
  96         // Estimate size overhead introduced by call instructions which
  97         // is higher than other instructions. Here 3 and 10 are magic
  98         // numbers that help one isolated test case from PR2067 without
  99         // negatively impacting measured benchmarks.
 100         if (isa<IntrinsicInst>(I))
 101           Size = Size + 3;
 102         else
 103           Size = Size + 10;
 104       } else {
 105         ++Size;
 106       }
 107
 108       // TODO: Ignore expressions derived from PHI and constants if inval of phi
 109       // is a constant, or if operation is associative.  This will get induction
 110       // variables.
 111     }
 112   }
 113
 114   return Size;
 115 }
 116
 117 bool LoopUnroll::runOnLoop(Loop *L, LPPassManager &LPM) {
 118   assert(L->isLCSSAForm());
 119   LoopInfo *LI = &getAnalysis<LoopInfo>();
 120
 121   BasicBlock *Header = L->getHeader();
 122   DEBUG(errs() << "Loop Unroll: F[" << Header->getParent()->getName()
 123         << "] Loop %" << Header->getName() << "\n");
 124   (void)Header;
 125
 126   // Find trip count
 127   unsigned TripCount = L->getSmallConstantTripCount();
 128   unsigned Count = UnrollCount;
 129
 130   // Automatically select an unroll count.
 131   if (Count == 0) {
 132     // Conservative heuristic: if we know the trip count, see if we can
 133     // completely unroll (subject to the threshold, checked below); otherwise
 134     // try to find greatest modulo of the trip count which is still under
 135     // threshold value.
 136     if (TripCount != 0) {
 137       Count = TripCount;
 138     } else {
 139       return false;
 140     }
 141   }
 142
 143   // Enforce the threshold.
 144   if (UnrollThreshold != NoThreshold) {
 145     unsigned LoopSize = ApproximateLoopSize(L);
 146     DOUT << "  Loop Size = " << LoopSize << "\n";
 147     uint64_t Size = (uint64_t)LoopSize*Count;
 148     if (TripCount != 1 && Size > UnrollThreshold) {
 149       DOUT << "  Too large to fully unroll with count: " << Count
 150            << " because size: " << Size << ">" << UnrollThreshold << "\n";
 151       if (UnrollAllowPartial) {
 152         // Reduce unroll count to be modulo of TripCount for partial unrolling
 153         Count = UnrollThreshold / LoopSize;
 154         while (Count != 0 && TripCount%Count != 0) {
 155           Count--;
 156         }
 157         if (Count < 2) {
 158           DOUT << "  could not unroll partially\n";
 159           return false;
 160         } else {
 161           DOUT << "  partially unrolling with count: " << Count << "\n";
 162         }
 163       } else {
 164         DOUT << "  will not try to unroll partially because "
 165              << "-unroll-allow-partial not given\n";
 166         return false;
 167       }
 168     }
 169   }
 170
 171   // Unroll the loop.
 172   Function *F = L->getHeader()->getParent();
 173   if (!UnrollLoop(L, Count, LI, &LPM))
 174     return false;
 175
 176   // FIXME: Reconstruct dom info, because it is not preserved properly.
 177   DominatorTree *DT = getAnalysisIfAvailable<DominatorTree>();
 178   if (DT) {
 179     DT->runOnFunction(*F);
 180     DominanceFrontier *DF = getAnalysisIfAvailable<DominanceFrontier>();
 181     if (DF)
 182       DF->runOnFunction(*F);
 183   }
 184   return true;
 185 }