LoopUnrollPass.cpp revision bbf81d88116d23fb0776412b5916f7d0b8b3ca7e 
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/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
26using namespace llvm;
27
28static cl::opt<unsigned>
29UnrollThreshold("unroll-threshold", cl::init(100), cl::Hidden,
30  cl::desc("The cut-off point for automatic loop unrolling"));
31
32static cl::opt<unsigned>
33UnrollCount("unroll-count", cl::init(0), cl::Hidden,
34  cl::desc("Use this unroll count for all loops, for testing purposes"));
35
36static cl::opt<bool>
37UnrollAllowPartial("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
41namespace {
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
73char LoopUnroll::ID = 0;
74static RegisterPass<LoopUnroll> X("loop-unroll", "Unroll loops");
75
76Pass *llvm::createLoopUnrollPass() { return new LoopUnroll(); }
77
78/// ApproximateLoopSize - Approximate the size of the loop.
79static unsigned ApproximateLoopSize(const Loop *L, unsigned &NumCalls) {
80  CodeMetrics Metrics;
81  for (Loop::block_iterator I = L->block_begin(), E = L->block_end();
82       I != E; ++I)
83    Metrics.analyzeBasicBlock(*I);
84  NumCalls = Metrics.NumCalls;
85  return Metrics.NumInsts;
86}
87
88bool LoopUnroll::runOnLoop(Loop *L, LPPassManager &LPM) {
89  LoopInfo *LI = &getAnalysis<LoopInfo>();
90
91  BasicBlock *Header = L->getHeader();
92  DEBUG(dbgs() << "Loop Unroll: F[" << Header->getParent()->getName()
93        << "] Loop %" << Header->getName() << "\n");
94  (void)Header;
95
96  // Find trip count
97  unsigned TripCount = L->getSmallConstantTripCount();
98  unsigned Count = UnrollCount;
99
100  // Automatically select an unroll count.
101  if (Count == 0) {
102    // Conservative heuristic: if we know the trip count, see if we can
103    // completely unroll (subject to the threshold, checked below); otherwise
104    // try to find greatest modulo of the trip count which is still under
105    // threshold value.
106    if (TripCount == 0)
107      return false;
108    Count = TripCount;
109  }
110
111  // Enforce the threshold.
112  if (UnrollThreshold != NoThreshold) {
113    unsigned NumCalls;
114    unsigned LoopSize = ApproximateLoopSize(L, NumCalls);
115    DEBUG(dbgs() << "  Loop Size = " << LoopSize << "\n");
116    if (NumCalls != 0) {
117      DEBUG(dbgs() << "  Not unrolling loop with function calls.\n");
118      return false;
119    }
120    uint64_t Size = (uint64_t)LoopSize*Count;
121    if (TripCount != 1 && Size > UnrollThreshold) {
122      DEBUG(dbgs() << "  Too large to fully unroll with count: " << Count
123            << " because size: " << Size << ">" << UnrollThreshold << "\n");
124      if (!UnrollAllowPartial) {
125        DEBUG(dbgs() << "  will not try to unroll partially because "
126              << "-unroll-allow-partial not given\n");
127        return false;
128      }
129      // Reduce unroll count to be modulo of TripCount for partial unrolling
130      Count = UnrollThreshold / LoopSize;
131      while (Count != 0 && TripCount%Count != 0) {
132        Count--;
133      }
134      if (Count < 2) {
135        DEBUG(dbgs() << "  could not unroll partially\n");
136        return false;
137      }
138      DEBUG(dbgs() << "  partially unrolling with count: " << Count << "\n");
139    }
140  }
141
142  // Unroll the loop.
143  Function *F = L->getHeader()->getParent();
144  if (!UnrollLoop(L, Count, LI, &LPM))
145    return false;
146
147  // FIXME: Reconstruct dom info, because it is not preserved properly.
148  DominatorTree *DT = getAnalysisIfAvailable<DominatorTree>();
149  if (DT) {
150    DT->runOnFunction(*F);
151    DominanceFrontier *DF = getAnalysisIfAvailable<DominanceFrontier>();
152    if (DF)
153      DF->runOnFunction(*F);
154  }
155  return true;
156}
157