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/Transforms/Scalar.h" 17 #include "llvm/Analysis/CodeMetrics.h" 18 #include "llvm/Analysis/LoopPass.h" 19 #include "llvm/Analysis/ScalarEvolution.h" 20 #include "llvm/DataLayout.h" 21 #include "llvm/IntrinsicInst.h" 22 #include "llvm/Support/CommandLine.h" 23 #include "llvm/Support/Debug.h" 24 #include "llvm/Support/raw_ostream.h" 25 #include "llvm/Transforms/Utils/UnrollLoop.h" 26 #include <climits> 27 28 using namespace llvm; 29 30 static cl::opt<unsigned> 31 UnrollThreshold("unroll-threshold", cl::init(150), cl::Hidden, 32 cl::desc("The cut-off point for automatic loop unrolling")); 33 34 static cl::opt<unsigned> 35 UnrollCount("unroll-count", cl::init(0), cl::Hidden, 36 cl::desc("Use this unroll count for all loops, for testing purposes")); 37 38 static cl::opt<bool> 39 UnrollAllowPartial("unroll-allow-partial", cl::init(false), cl::Hidden, 40 cl::desc("Allows loops to be partially unrolled until " 41 "-unroll-threshold loop size is reached.")); 42 43 static cl::opt<bool> 44 UnrollRuntime("unroll-runtime", cl::ZeroOrMore, cl::init(false), cl::Hidden, 45 cl::desc("Unroll loops with run-time trip counts")); 46 47 namespace { 48 class LoopUnroll : public LoopPass { 49 public: 50 static char ID; // Pass ID, replacement for typeid 51 LoopUnroll(int T = -1, int C = -1, int P = -1) : LoopPass(ID) { 52 CurrentThreshold = (T == -1) ? UnrollThreshold : unsigned(T); 53 CurrentCount = (C == -1) ? UnrollCount : unsigned(C); 54 CurrentAllowPartial = (P == -1) ? UnrollAllowPartial : (bool)P; 55 56 UserThreshold = (T != -1) || (UnrollThreshold.getNumOccurrences() > 0); 57 58 initializeLoopUnrollPass(*PassRegistry::getPassRegistry()); 59 } 60 61 /// A magic value for use with the Threshold parameter to indicate 62 /// that the loop unroll should be performed regardless of how much 63 /// code expansion would result. 64 static const unsigned NoThreshold = UINT_MAX; 65 66 // Threshold to use when optsize is specified (and there is no 67 // explicit -unroll-threshold). 68 static const unsigned OptSizeUnrollThreshold = 50; 69 70 // Default unroll count for loops with run-time trip count if 71 // -unroll-count is not set 72 static const unsigned UnrollRuntimeCount = 8; 73 74 unsigned CurrentCount; 75 unsigned CurrentThreshold; 76 bool CurrentAllowPartial; 77 bool UserThreshold; // CurrentThreshold is user-specified. 78 79 bool runOnLoop(Loop *L, LPPassManager &LPM); 80 81 /// This transformation requires natural loop information & requires that 82 /// loop preheaders be inserted into the CFG... 83 /// 84 virtual void getAnalysisUsage(AnalysisUsage &AU) const { 85 AU.addRequired<LoopInfo>(); 86 AU.addPreserved<LoopInfo>(); 87 AU.addRequiredID(LoopSimplifyID); 88 AU.addPreservedID(LoopSimplifyID); 89 AU.addRequiredID(LCSSAID); 90 AU.addPreservedID(LCSSAID); 91 AU.addRequired<ScalarEvolution>(); 92 AU.addPreserved<ScalarEvolution>(); 93 // FIXME: Loop unroll requires LCSSA. And LCSSA requires dom info. 94 // If loop unroll does not preserve dom info then LCSSA pass on next 95 // loop will receive invalid dom info. 96 // For now, recreate dom info, if loop is unrolled. 97 AU.addPreserved<DominatorTree>(); 98 } 99 }; 100 } 101 102 char LoopUnroll::ID = 0; 103 INITIALIZE_PASS_BEGIN(LoopUnroll, "loop-unroll", "Unroll loops", false, false) 104 INITIALIZE_PASS_DEPENDENCY(LoopInfo) 105 INITIALIZE_PASS_DEPENDENCY(LoopSimplify) 106 INITIALIZE_PASS_DEPENDENCY(LCSSA) 107 INITIALIZE_PASS_DEPENDENCY(ScalarEvolution) 108 INITIALIZE_PASS_END(LoopUnroll, "loop-unroll", "Unroll loops", false, false) 109 110 Pass *llvm::createLoopUnrollPass(int Threshold, int Count, int AllowPartial) { 111 return new LoopUnroll(Threshold, Count, AllowPartial); 112 } 113 114 /// ApproximateLoopSize - Approximate the size of the loop. 115 static unsigned ApproximateLoopSize(const Loop *L, unsigned &NumCalls, 116 bool &NotDuplicatable, const DataLayout *TD) { 117 CodeMetrics Metrics; 118 for (Loop::block_iterator I = L->block_begin(), E = L->block_end(); 119 I != E; ++I) 120 Metrics.analyzeBasicBlock(*I, TD); 121 NumCalls = Metrics.NumInlineCandidates; 122 NotDuplicatable = Metrics.notDuplicatable; 123 124 unsigned LoopSize = Metrics.NumInsts; 125 126 // Don't allow an estimate of size zero. This would allows unrolling of loops 127 // with huge iteration counts, which is a compile time problem even if it's 128 // not a problem for code quality. 129 if (LoopSize == 0) LoopSize = 1; 130 131 return LoopSize; 132 } 133 134 bool LoopUnroll::runOnLoop(Loop *L, LPPassManager &LPM) { 135 LoopInfo *LI = &getAnalysis<LoopInfo>(); 136 ScalarEvolution *SE = &getAnalysis<ScalarEvolution>(); 137 138 BasicBlock *Header = L->getHeader(); 139 DEBUG(dbgs() << "Loop Unroll: F[" << Header->getParent()->getName() 140 << "] Loop %" << Header->getName() << "\n"); 141 (void)Header; 142 143 // Determine the current unrolling threshold. While this is normally set 144 // from UnrollThreshold, it is overridden to a smaller value if the current 145 // function is marked as optimize-for-size, and the unroll threshold was 146 // not user specified. 147 unsigned Threshold = CurrentThreshold; 148 if (!UserThreshold && 149 Header->getParent()->getAttributes(). 150 hasAttribute(AttributeSet::FunctionIndex, 151 Attribute::OptimizeForSize)) 152 Threshold = OptSizeUnrollThreshold; 153 154 // Find trip count and trip multiple if count is not available 155 unsigned TripCount = 0; 156 unsigned TripMultiple = 1; 157 // Find "latch trip count". UnrollLoop assumes that control cannot exit 158 // via the loop latch on any iteration prior to TripCount. The loop may exit 159 // early via an earlier branch. 160 BasicBlock *LatchBlock = L->getLoopLatch(); 161 if (LatchBlock) { 162 TripCount = SE->getSmallConstantTripCount(L, LatchBlock); 163 TripMultiple = SE->getSmallConstantTripMultiple(L, LatchBlock); 164 } 165 // Use a default unroll-count if the user doesn't specify a value 166 // and the trip count is a run-time value. The default is different 167 // for run-time or compile-time trip count loops. 168 unsigned Count = CurrentCount; 169 if (UnrollRuntime && CurrentCount == 0 && TripCount == 0) 170 Count = UnrollRuntimeCount; 171 172 if (Count == 0) { 173 // Conservative heuristic: if we know the trip count, see if we can 174 // completely unroll (subject to the threshold, checked below); otherwise 175 // try to find greatest modulo of the trip count which is still under 176 // threshold value. 177 if (TripCount == 0) 178 return false; 179 Count = TripCount; 180 } 181 182 // Enforce the threshold. 183 if (Threshold != NoThreshold) { 184 const DataLayout *TD = getAnalysisIfAvailable<DataLayout>(); 185 unsigned NumInlineCandidates; 186 bool notDuplicatable; 187 unsigned LoopSize = ApproximateLoopSize(L, NumInlineCandidates, 188 notDuplicatable, TD); 189 DEBUG(dbgs() << " Loop Size = " << LoopSize << "\n"); 190 if (notDuplicatable) { 191 DEBUG(dbgs() << " Not unrolling loop which contains non duplicatable" 192 << " instructions.\n"); 193 return false; 194 } 195 if (NumInlineCandidates != 0) { 196 DEBUG(dbgs() << " Not unrolling loop with inlinable calls.\n"); 197 return false; 198 } 199 uint64_t Size = (uint64_t)LoopSize*Count; 200 if (TripCount != 1 && Size > Threshold) { 201 DEBUG(dbgs() << " Too large to fully unroll with count: " << Count 202 << " because size: " << Size << ">" << Threshold << "\n"); 203 if (!CurrentAllowPartial && !(UnrollRuntime && TripCount == 0)) { 204 DEBUG(dbgs() << " will not try to unroll partially because " 205 << "-unroll-allow-partial not given\n"); 206 return false; 207 } 208 if (TripCount) { 209 // Reduce unroll count to be modulo of TripCount for partial unrolling 210 Count = Threshold / LoopSize; 211 while (Count != 0 && TripCount%Count != 0) 212 Count--; 213 } 214 else if (UnrollRuntime) { 215 // Reduce unroll count to be a lower power-of-two value 216 while (Count != 0 && Size > Threshold) { 217 Count >>= 1; 218 Size = LoopSize*Count; 219 } 220 } 221 if (Count < 2) { 222 DEBUG(dbgs() << " could not unroll partially\n"); 223 return false; 224 } 225 DEBUG(dbgs() << " partially unrolling with count: " << Count << "\n"); 226 } 227 } 228 229 // Unroll the loop. 230 if (!UnrollLoop(L, Count, TripCount, UnrollRuntime, TripMultiple, LI, &LPM)) 231 return false; 232 233 return true; 234 } 235