lib/Transforms/Scalar/LoopUnrollPass.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/LoopPass.h"
  19 #include "llvm/Analysis/InlineCost.h"
  20 #include "llvm/Analysis/ScalarEvolution.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(150), 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 LoopUnroll : public LoopPass {
  44   public:
  45     static char ID; // Pass ID, replacement for typeid
  46     LoopUnroll() : LoopPass(ID) {
  47       initializeLoopUnrollPass(*PassRegistry::getPassRegistry());
  48     }
  49
  50     /// A magic value for use with the Threshold parameter to indicate
  51     /// that the loop unroll should be performed regardless of how much
  52     /// code expansion would result.
  53     static const unsigned NoThreshold = UINT_MAX;
  54
  55     // Threshold to use when optsize is specified (and there is no
  56     // explicit -unroll-threshold).
  57     static const unsigned OptSizeUnrollThreshold = 50;
  58
  59     unsigned CurrentThreshold;
  60
  61     bool runOnLoop(Loop *L, LPPassManager &LPM);
  62
  63     /// This transformation requires natural loop information & requires that
  64     /// loop preheaders be inserted into the CFG...
  65     ///
  66     virtual void getAnalysisUsage(AnalysisUsage &AU) const {
  67       AU.addRequired<LoopInfo>();
  68       AU.addPreserved<LoopInfo>();
  69       AU.addRequiredID(LoopSimplifyID);
  70       AU.addPreservedID(LoopSimplifyID);
  71       AU.addRequiredID(LCSSAID);
  72       AU.addPreservedID(LCSSAID);
  73       AU.addPreserved<ScalarEvolution>();
  74       // FIXME: Loop unroll requires LCSSA. And LCSSA requires dom info.
  75       // If loop unroll does not preserve dom info then LCSSA pass on next
  76       // loop will receive invalid dom info.
  77       // For now, recreate dom info, if loop is unrolled.
  78       AU.addPreserved<DominatorTree>();
  79     }
  80   };
  81 }
  82
  83 char LoopUnroll::ID = 0;
  84 INITIALIZE_PASS_BEGIN(LoopUnroll, "loop-unroll", "Unroll loops", false, false)
  85 INITIALIZE_PASS_DEPENDENCY(LoopInfo)
  86 INITIALIZE_PASS_DEPENDENCY(LoopSimplify)
  87 INITIALIZE_PASS_DEPENDENCY(LCSSA)
  88 INITIALIZE_PASS_END(LoopUnroll, "loop-unroll", "Unroll loops", false, false)
  89
  90 Pass *llvm::createLoopUnrollPass() { return new LoopUnroll(); }
  91
  92 /// ApproximateLoopSize - Approximate the size of the loop.
  93 static unsigned ApproximateLoopSize(const Loop *L, unsigned &NumCalls) {
  94   CodeMetrics Metrics;
  95   for (Loop::block_iterator I = L->block_begin(), E = L->block_end();
  96        I != E; ++I)
  97     Metrics.analyzeBasicBlock(*I);
  98   NumCalls = Metrics.NumInlineCandidates;
  99
 100   unsigned LoopSize = Metrics.NumInsts;
 101
 102   // If we can identify the induction variable, we know that it will become
 103   // constant when we unroll the loop, so factor that into our loop size
 104   // estimate.
 105   // FIXME: We have to divide by InlineConstants::InstrCost because the
 106   // measure returned by CountCodeReductionForConstant is not an instruction
 107   // count, but rather a weight as defined by InlineConstants.  It would
 108   // probably be a good idea to standardize on a single weighting scheme by
 109   // pushing more of the logic for weighting into CodeMetrics.
 110   if (PHINode *IndVar = L->getCanonicalInductionVariable()) {
 111     unsigned SizeDecrease = Metrics.CountCodeReductionForConstant(IndVar);
 112     // NOTE: Because SizeDecrease is a fuzzy estimate, we don't want to allow
 113     // it to totally negate the cost of unrolling a loop.
 114     SizeDecrease = SizeDecrease > LoopSize / 2 ? LoopSize / 2 : SizeDecrease;
 115   }
 116
 117   // Don't allow an estimate of size zero.  This would allows unrolling of loops
 118   // with huge iteration counts, which is a compile time problem even if it's
 119   // not a problem for code quality.
 120   if (LoopSize == 0) LoopSize = 1;
 121
 122   return LoopSize;
 123 }
 124
 125 bool LoopUnroll::runOnLoop(Loop *L, LPPassManager &LPM) {
 126
 127   LoopInfo *LI = &getAnalysis<LoopInfo>();
 128
 129   BasicBlock *Header = L->getHeader();
 130   DEBUG(dbgs() << "Loop Unroll: F[" << Header->getParent()->getName()
 131         << "] Loop %" << Header->getName() << "\n");
 132   (void)Header;
 133
 134   // Determine the current unrolling threshold.  While this is normally set
 135   // from UnrollThreshold, it is overridden to a smaller value if the current
 136   // function is marked as optimize-for-size, and the unroll threshold was
 137   // not user specified.
 138   CurrentThreshold = UnrollThreshold;
 139   if (Header->getParent()->hasFnAttr(Attribute::OptimizeForSize) &&
 140       UnrollThreshold.getNumOccurrences() == 0)
 141     CurrentThreshold = OptSizeUnrollThreshold;
 142
 143   // Find trip count
 144   unsigned TripCount = L->getSmallConstantTripCount();
 145   unsigned Count = UnrollCount;
 146
 147   // Automatically select an unroll count.
 148   if (Count == 0) {
 149     // Conservative heuristic: if we know the trip count, see if we can
 150     // completely unroll (subject to the threshold, checked below); otherwise
 151     // try to find greatest modulo of the trip count which is still under
 152     // threshold value.
 153     if (TripCount == 0)
 154       return false;
 155     Count = TripCount;
 156   }
 157
 158   // Enforce the threshold.
 159   if (CurrentThreshold != NoThreshold) {
 160     unsigned NumInlineCandidates;
 161     unsigned LoopSize = ApproximateLoopSize(L, NumInlineCandidates);
 162     DEBUG(dbgs() << "  Loop Size = " << LoopSize << "\n");
 163     if (NumInlineCandidates != 0) {
 164       DEBUG(dbgs() << "  Not unrolling loop with inlinable calls.\n");
 165       return false;
 166     }
 167     uint64_t Size = (uint64_t)LoopSize*Count;
 168     if (TripCount != 1 && Size > CurrentThreshold) {
 169       DEBUG(dbgs() << "  Too large to fully unroll with count: " << Count
 170             << " because size: " << Size << ">" << CurrentThreshold << "\n");
 171       if (!UnrollAllowPartial) {
 172         DEBUG(dbgs() << "  will not try to unroll partially because "
 173               << "-unroll-allow-partial not given\n");
 174         return false;
 175       }
 176       // Reduce unroll count to be modulo of TripCount for partial unrolling
 177       Count = CurrentThreshold / LoopSize;
 178       while (Count != 0 && TripCount%Count != 0) {
 179         Count--;
 180       }
 181       if (Count < 2) {
 182         DEBUG(dbgs() << "  could not unroll partially\n");
 183         return false;
 184       }
 185       DEBUG(dbgs() << "  partially unrolling with count: " << Count << "\n");
 186     }
 187   }
 188
 189   // Unroll the loop.
 190   Function *F = L->getHeader()->getParent();
 191   if (!UnrollLoop(L, Count, LI, &LPM))
 192     return false;
 193
 194   // FIXME: Reconstruct dom info, because it is not preserved properly.
 195   if (DominatorTree *DT = getAnalysisIfAvailable<DominatorTree>())
 196     DT->runOnFunction(*F);
 197   return true;
 198 }