LoopUnrollPass.cpp revision ce665bd2e2b581ab0858d1afe359192bac96b868 
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
27using namespace llvm;
28
29static cl::opt<unsigned>
30UnrollThreshold("unroll-threshold", cl::init(150), cl::Hidden,
31  cl::desc("The cut-off point for automatic loop unrolling"));
32
33static cl::opt<unsigned>
34UnrollCount("unroll-count", cl::init(0), cl::Hidden,
35  cl::desc("Use this unroll count for all loops, for testing purposes"));
36
37static cl::opt<bool>
38UnrollAllowPartial("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
42namespace {
43  class 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    // Threshold to use when optsize is specified (and there is no
54    // explicit -unroll-threshold).
55    static const unsigned OptSizeUnrollThreshold = 50;
56
57    unsigned CurrentThreshold;
58
59    bool runOnLoop(Loop *L, LPPassManager &LPM);
60
61    /// This transformation requires natural loop information & requires that
62    /// loop preheaders be inserted into the CFG...
63    ///
64    virtual void getAnalysisUsage(AnalysisUsage &AU) const {
65      AU.addRequired<LoopInfo>();
66      AU.addPreserved<LoopInfo>();
67      AU.addRequiredID(LoopSimplifyID);
68      AU.addPreservedID(LoopSimplifyID);
69      AU.addRequiredID(LCSSAID);
70      AU.addPreservedID(LCSSAID);
71      AU.addPreserved<ScalarEvolution>();
72      // FIXME: Loop unroll requires LCSSA. And LCSSA requires dom info.
73      // If loop unroll does not preserve dom info then LCSSA pass on next
74      // loop will receive invalid dom info.
75      // For now, recreate dom info, if loop is unrolled.
76      AU.addPreserved<DominatorTree>();
77    }
78  };
79}
80
81char LoopUnroll::ID = 0;
82INITIALIZE_PASS(LoopUnroll, "loop-unroll", "Unroll loops", false, false)
83
84Pass *llvm::createLoopUnrollPass() { return new LoopUnroll(); }
85
86/// ApproximateLoopSize - Approximate the size of the loop.
87static unsigned ApproximateLoopSize(const Loop *L, unsigned &NumCalls) {
88  CodeMetrics Metrics;
89  for (Loop::block_iterator I = L->block_begin(), E = L->block_end();
90       I != E; ++I)
91    Metrics.analyzeBasicBlock(*I);
92  NumCalls = Metrics.NumInlineCandidates;
93
94  unsigned LoopSize = Metrics.NumInsts;
95
96  // If we can identify the induction variable, we know that it will become
97  // constant when we unroll the loop, so factor that into our loop size
98  // estimate.
99  // FIXME: We have to divide by InlineConstants::InstrCost because the
100  // measure returned by CountCodeReductionForConstant is not an instruction
101  // count, but rather a weight as defined by InlineConstants.  It would
102  // probably be a good idea to standardize on a single weighting scheme by
103  // pushing more of the logic for weighting into CodeMetrics.
104  if (PHINode *IndVar = L->getCanonicalInductionVariable()) {
105    unsigned SizeDecrease = Metrics.CountCodeReductionForConstant(IndVar);
106    // NOTE: Because SizeDecrease is a fuzzy estimate, we don't want to allow
107    // it to totally negate the cost of unrolling a loop.
108    SizeDecrease = SizeDecrease > LoopSize / 2 ? LoopSize / 2 : SizeDecrease;
109  }
110
111  // Don't allow an estimate of size zero.  This would allows unrolling of loops
112  // with huge iteration counts, which is a compile time problem even if it's
113  // not a problem for code quality.
114  if (LoopSize == 0) LoopSize = 1;
115
116  return LoopSize;
117}
118
119bool LoopUnroll::runOnLoop(Loop *L, LPPassManager &LPM) {
120
121  LoopInfo *LI = &getAnalysis<LoopInfo>();
122
123  BasicBlock *Header = L->getHeader();
124  DEBUG(dbgs() << "Loop Unroll: F[" << Header->getParent()->getName()
125        << "] Loop %" << Header->getName() << "\n");
126  (void)Header;
127
128  // Determine the current unrolling threshold.  While this is normally set
129  // from UnrollThreshold, it is overridden to a smaller value if the current
130  // function is marked as optimize-for-size, and the unroll threshold was
131  // not user specified.
132  CurrentThreshold = UnrollThreshold;
133  if (Header->getParent()->hasFnAttr(Attribute::OptimizeForSize) &&
134      UnrollThreshold.getNumOccurrences() == 0)
135    CurrentThreshold = OptSizeUnrollThreshold;
136
137  // Find trip count
138  unsigned TripCount = L->getSmallConstantTripCount();
139  unsigned Count = UnrollCount;
140
141  // Automatically select an unroll count.
142  if (Count == 0) {
143    // Conservative heuristic: if we know the trip count, see if we can
144    // completely unroll (subject to the threshold, checked below); otherwise
145    // try to find greatest modulo of the trip count which is still under
146    // threshold value.
147    if (TripCount == 0)
148      return false;
149    Count = TripCount;
150  }
151
152  // Enforce the threshold.
153  if (CurrentThreshold != NoThreshold) {
154    unsigned NumInlineCandidates;
155    unsigned LoopSize = ApproximateLoopSize(L, NumInlineCandidates);
156    DEBUG(dbgs() << "  Loop Size = " << LoopSize << "\n");
157    if (NumInlineCandidates != 0) {
158      DEBUG(dbgs() << "  Not unrolling loop with inlinable calls.\n");
159      return false;
160    }
161    uint64_t Size = (uint64_t)LoopSize*Count;
162    if (TripCount != 1 && Size > CurrentThreshold) {
163      DEBUG(dbgs() << "  Too large to fully unroll with count: " << Count
164            << " because size: " << Size << ">" << CurrentThreshold << "\n");
165      if (!UnrollAllowPartial) {
166        DEBUG(dbgs() << "  will not try to unroll partially because "
167              << "-unroll-allow-partial not given\n");
168        return false;
169      }
170      // Reduce unroll count to be modulo of TripCount for partial unrolling
171      Count = CurrentThreshold / LoopSize;
172      while (Count != 0 && TripCount%Count != 0) {
173        Count--;
174      }
175      if (Count < 2) {
176        DEBUG(dbgs() << "  could not unroll partially\n");
177        return false;
178      }
179      DEBUG(dbgs() << "  partially unrolling with count: " << Count << "\n");
180    }
181  }
182
183  // Unroll the loop.
184  Function *F = L->getHeader()->getParent();
185  if (!UnrollLoop(L, Count, LI, &LPM))
186    return false;
187
188  // FIXME: Reconstruct dom info, because it is not preserved properly.
189  if (DominatorTree *DT = getAnalysisIfAvailable<DominatorTree>())
190    DT->runOnFunction(*F);
191  return true;
192}
193