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/Analysis/TargetTransformInfo.h" 21 #include "llvm/IR/DataLayout.h" 22 #include "llvm/IR/IntrinsicInst.h" 23 #include "llvm/Support/CommandLine.h" 24 #include "llvm/Support/Debug.h" 25 #include "llvm/Support/raw_ostream.h" 26 #include "llvm/Transforms/Utils/UnrollLoop.h" 27 #include <climits> 28 29 using namespace llvm; 30 31 static cl::opt<unsigned> 32 UnrollThreshold("unroll-threshold", cl::init(150), cl::Hidden, 33 cl::desc("The cut-off point for automatic loop unrolling")); 34 35 static cl::opt<unsigned> 36 UnrollCount("unroll-count", cl::init(0), cl::Hidden, 37 cl::desc("Use this unroll count for all loops, for testing purposes")); 38 39 static cl::opt<bool> 40 UnrollAllowPartial("unroll-allow-partial", cl::init(false), cl::Hidden, 41 cl::desc("Allows loops to be partially unrolled until " 42 "-unroll-threshold loop size is reached.")); 43 44 static cl::opt<bool> 45 UnrollRuntime("unroll-runtime", cl::ZeroOrMore, cl::init(false), cl::Hidden, 46 cl::desc("Unroll loops with run-time trip counts")); 47 48 namespace { 49 class LoopUnroll : public LoopPass { 50 public: 51 static char ID; // Pass ID, replacement for typeid 52 LoopUnroll(int T = -1, int C = -1, int P = -1) : LoopPass(ID) { 53 CurrentThreshold = (T == -1) ? UnrollThreshold : unsigned(T); 54 CurrentCount = (C == -1) ? UnrollCount : unsigned(C); 55 CurrentAllowPartial = (P == -1) ? UnrollAllowPartial : (bool)P; 56 57 UserThreshold = (T != -1) || (UnrollThreshold.getNumOccurrences() > 0); 58 UserAllowPartial = (P != -1) || 59 (UnrollAllowPartial.getNumOccurrences() > 0); 60 61 initializeLoopUnrollPass(*PassRegistry::getPassRegistry()); 62 } 63 64 /// A magic value for use with the Threshold parameter to indicate 65 /// that the loop unroll should be performed regardless of how much 66 /// code expansion would result. 67 static const unsigned NoThreshold = UINT_MAX; 68 69 // Threshold to use when optsize is specified (and there is no 70 // explicit -unroll-threshold). 71 static const unsigned OptSizeUnrollThreshold = 50; 72 73 // Default unroll count for loops with run-time trip count if 74 // -unroll-count is not set 75 static const unsigned UnrollRuntimeCount = 8; 76 77 unsigned CurrentCount; 78 unsigned CurrentThreshold; 79 bool CurrentAllowPartial; 80 bool UserThreshold; // CurrentThreshold is user-specified. 81 bool UserAllowPartial; // CurrentAllowPartial is user-specified. 82 83 bool runOnLoop(Loop *L, LPPassManager &LPM); 84 85 /// This transformation requires natural loop information & requires that 86 /// loop preheaders be inserted into the CFG... 87 /// 88 virtual void getAnalysisUsage(AnalysisUsage &AU) const { 89 AU.addRequired<LoopInfo>(); 90 AU.addPreserved<LoopInfo>(); 91 AU.addRequiredID(LoopSimplifyID); 92 AU.addPreservedID(LoopSimplifyID); 93 AU.addRequiredID(LCSSAID); 94 AU.addPreservedID(LCSSAID); 95 AU.addRequired<ScalarEvolution>(); 96 AU.addPreserved<ScalarEvolution>(); 97 AU.addRequired<TargetTransformInfo>(); 98 // FIXME: Loop unroll requires LCSSA. And LCSSA requires dom info. 99 // If loop unroll does not preserve dom info then LCSSA pass on next 100 // loop will receive invalid dom info. 101 // For now, recreate dom info, if loop is unrolled. 102 AU.addPreserved<DominatorTree>(); 103 } 104 }; 105 } 106 107 char LoopUnroll::ID = 0; 108 INITIALIZE_PASS_BEGIN(LoopUnroll, "loop-unroll", "Unroll loops", false, false) 109 INITIALIZE_AG_DEPENDENCY(TargetTransformInfo) 110 INITIALIZE_PASS_DEPENDENCY(LoopInfo) 111 INITIALIZE_PASS_DEPENDENCY(LoopSimplify) 112 INITIALIZE_PASS_DEPENDENCY(LCSSA) 113 INITIALIZE_PASS_DEPENDENCY(ScalarEvolution) 114 INITIALIZE_PASS_END(LoopUnroll, "loop-unroll", "Unroll loops", false, false) 115 116 Pass *llvm::createLoopUnrollPass(int Threshold, int Count, int AllowPartial) { 117 return new LoopUnroll(Threshold, Count, AllowPartial); 118 } 119 120 /// ApproximateLoopSize - Approximate the size of the loop. 121 static unsigned ApproximateLoopSize(const Loop *L, unsigned &NumCalls, 122 bool &NotDuplicatable, 123 const TargetTransformInfo &TTI) { 124 CodeMetrics Metrics; 125 for (Loop::block_iterator I = L->block_begin(), E = L->block_end(); 126 I != E; ++I) 127 Metrics.analyzeBasicBlock(*I, TTI); 128 NumCalls = Metrics.NumInlineCandidates; 129 NotDuplicatable = Metrics.notDuplicatable; 130 131 unsigned LoopSize = Metrics.NumInsts; 132 133 // Don't allow an estimate of size zero. This would allows unrolling of loops 134 // with huge iteration counts, which is a compile time problem even if it's 135 // not a problem for code quality. 136 if (LoopSize == 0) LoopSize = 1; 137 138 return LoopSize; 139 } 140 141 bool LoopUnroll::runOnLoop(Loop *L, LPPassManager &LPM) { 142 LoopInfo *LI = &getAnalysis<LoopInfo>(); 143 ScalarEvolution *SE = &getAnalysis<ScalarEvolution>(); 144 const TargetTransformInfo &TTI = getAnalysis<TargetTransformInfo>(); 145 146 BasicBlock *Header = L->getHeader(); 147 DEBUG(dbgs() << "Loop Unroll: F[" << Header->getParent()->getName() 148 << "] Loop %" << Header->getName() << "\n"); 149 (void)Header; 150 151 TargetTransformInfo::UnrollingPreferences UP; 152 bool HasUP = TTI.getUnrollingPreferences(UP); 153 154 // Determine the current unrolling threshold. While this is normally set 155 // from UnrollThreshold, it is overridden to a smaller value if the current 156 // function is marked as optimize-for-size, and the unroll threshold was 157 // not user specified. 158 unsigned Threshold = (HasUP && !UserThreshold) ? UP.Threshold : 159 CurrentThreshold; 160 if (!UserThreshold && 161 Header->getParent()->getAttributes(). 162 hasAttribute(AttributeSet::FunctionIndex, 163 Attribute::OptimizeForSize)) 164 Threshold = HasUP ? UP.OptSizeThreshold : OptSizeUnrollThreshold; 165 166 // Find trip count and trip multiple if count is not available 167 unsigned TripCount = 0; 168 unsigned TripMultiple = 1; 169 // Find "latch trip count". UnrollLoop assumes that control cannot exit 170 // via the loop latch on any iteration prior to TripCount. The loop may exit 171 // early via an earlier branch. 172 BasicBlock *LatchBlock = L->getLoopLatch(); 173 if (LatchBlock) { 174 TripCount = SE->getSmallConstantTripCount(L, LatchBlock); 175 TripMultiple = SE->getSmallConstantTripMultiple(L, LatchBlock); 176 } 177 // Use a default unroll-count if the user doesn't specify a value 178 // and the trip count is a run-time value. The default is different 179 // for run-time or compile-time trip count loops. 180 unsigned Count = CurrentCount; 181 if (UnrollRuntime && CurrentCount == 0 && TripCount == 0) 182 Count = UnrollRuntimeCount; 183 184 if (Count == 0) { 185 // Conservative heuristic: if we know the trip count, see if we can 186 // completely unroll (subject to the threshold, checked below); otherwise 187 // try to find greatest modulo of the trip count which is still under 188 // threshold value. 189 if (TripCount == 0) 190 return false; 191 Count = TripCount; 192 } 193 194 bool Runtime = (HasUP && UnrollRuntime.getNumOccurrences() == 0) ? 195 UP.Runtime : UnrollRuntime; 196 197 // Enforce the threshold. 198 if (Threshold != NoThreshold) { 199 unsigned NumInlineCandidates; 200 bool notDuplicatable; 201 unsigned LoopSize = ApproximateLoopSize(L, NumInlineCandidates, 202 notDuplicatable, TTI); 203 DEBUG(dbgs() << " Loop Size = " << LoopSize << "\n"); 204 if (notDuplicatable) { 205 DEBUG(dbgs() << " Not unrolling loop which contains non duplicatable" 206 << " instructions.\n"); 207 return false; 208 } 209 if (NumInlineCandidates != 0) { 210 DEBUG(dbgs() << " Not unrolling loop with inlinable calls.\n"); 211 return false; 212 } 213 uint64_t Size = (uint64_t)LoopSize*Count; 214 if (TripCount != 1 && Size > Threshold) { 215 DEBUG(dbgs() << " Too large to fully unroll with count: " << Count 216 << " because size: " << Size << ">" << Threshold << "\n"); 217 bool AllowPartial = (HasUP && !UserAllowPartial) ? UP.Partial : 218 CurrentAllowPartial; 219 if (!AllowPartial && !(Runtime && TripCount == 0)) { 220 DEBUG(dbgs() << " will not try to unroll partially because " 221 << "-unroll-allow-partial not given\n"); 222 return false; 223 } 224 if (TripCount) { 225 // Reduce unroll count to be modulo of TripCount for partial unrolling 226 Count = Threshold / LoopSize; 227 while (Count != 0 && TripCount%Count != 0) 228 Count--; 229 } 230 else if (Runtime) { 231 // Reduce unroll count to be a lower power-of-two value 232 while (Count != 0 && Size > Threshold) { 233 Count >>= 1; 234 Size = LoopSize*Count; 235 } 236 } 237 if (Count < 2) { 238 DEBUG(dbgs() << " could not unroll partially\n"); 239 return false; 240 } 241 DEBUG(dbgs() << " partially unrolling with count: " << Count << "\n"); 242 } 243 } 244 245 // Unroll the loop. 246 if (!UnrollLoop(L, Count, TripCount, Runtime, TripMultiple, LI, &LPM)) 247 return false; 248 249 return true; 250 } 251