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 #include "llvm/Transforms/Scalar.h" 16 #include "llvm/Analysis/CodeMetrics.h" 17 #include "llvm/Analysis/LoopPass.h" 18 #include "llvm/Analysis/ScalarEvolution.h" 19 #include "llvm/Analysis/TargetTransformInfo.h" 20 #include "llvm/IR/DataLayout.h" 21 #include "llvm/IR/DiagnosticInfo.h" 22 #include "llvm/IR/Dominators.h" 23 #include "llvm/IR/IntrinsicInst.h" 24 #include "llvm/IR/Metadata.h" 25 #include "llvm/Support/CommandLine.h" 26 #include "llvm/Support/Debug.h" 27 #include "llvm/Support/raw_ostream.h" 28 #include "llvm/Transforms/Utils/UnrollLoop.h" 29 #include <climits> 30 31 using namespace llvm; 32 33 #define DEBUG_TYPE "loop-unroll" 34 35 static cl::opt<unsigned> 36 UnrollThreshold("unroll-threshold", cl::init(150), cl::Hidden, 37 cl::desc("The cut-off point for automatic loop unrolling")); 38 39 static cl::opt<unsigned> 40 UnrollCount("unroll-count", cl::init(0), cl::Hidden, 41 cl::desc("Use this unroll count for all loops including those with " 42 "unroll_count pragma values, for testing purposes")); 43 44 static cl::opt<bool> 45 UnrollAllowPartial("unroll-allow-partial", cl::init(false), cl::Hidden, 46 cl::desc("Allows loops to be partially unrolled until " 47 "-unroll-threshold loop size is reached.")); 48 49 static cl::opt<bool> 50 UnrollRuntime("unroll-runtime", cl::ZeroOrMore, cl::init(false), cl::Hidden, 51 cl::desc("Unroll loops with run-time trip counts")); 52 53 static cl::opt<unsigned> 54 PragmaUnrollThreshold("pragma-unroll-threshold", cl::init(16 * 1024), cl::Hidden, 55 cl::desc("Unrolled size limit for loops with an unroll(enable) or " 56 "unroll_count pragma.")); 57 58 namespace { 59 class LoopUnroll : public LoopPass { 60 public: 61 static char ID; // Pass ID, replacement for typeid 62 LoopUnroll(int T = -1, int C = -1, int P = -1, int R = -1) : LoopPass(ID) { 63 CurrentThreshold = (T == -1) ? UnrollThreshold : unsigned(T); 64 CurrentCount = (C == -1) ? UnrollCount : unsigned(C); 65 CurrentAllowPartial = (P == -1) ? UnrollAllowPartial : (bool)P; 66 CurrentRuntime = (R == -1) ? UnrollRuntime : (bool)R; 67 68 UserThreshold = (T != -1) || (UnrollThreshold.getNumOccurrences() > 0); 69 UserAllowPartial = (P != -1) || 70 (UnrollAllowPartial.getNumOccurrences() > 0); 71 UserRuntime = (R != -1) || (UnrollRuntime.getNumOccurrences() > 0); 72 UserCount = (C != -1) || (UnrollCount.getNumOccurrences() > 0); 73 74 initializeLoopUnrollPass(*PassRegistry::getPassRegistry()); 75 } 76 77 /// A magic value for use with the Threshold parameter to indicate 78 /// that the loop unroll should be performed regardless of how much 79 /// code expansion would result. 80 static const unsigned NoThreshold = UINT_MAX; 81 82 // Threshold to use when optsize is specified (and there is no 83 // explicit -unroll-threshold). 84 static const unsigned OptSizeUnrollThreshold = 50; 85 86 // Default unroll count for loops with run-time trip count if 87 // -unroll-count is not set 88 static const unsigned UnrollRuntimeCount = 8; 89 90 unsigned CurrentCount; 91 unsigned CurrentThreshold; 92 bool CurrentAllowPartial; 93 bool CurrentRuntime; 94 bool UserCount; // CurrentCount is user-specified. 95 bool UserThreshold; // CurrentThreshold is user-specified. 96 bool UserAllowPartial; // CurrentAllowPartial is user-specified. 97 bool UserRuntime; // CurrentRuntime is user-specified. 98 99 bool runOnLoop(Loop *L, LPPassManager &LPM) override; 100 101 /// This transformation requires natural loop information & requires that 102 /// loop preheaders be inserted into the CFG... 103 /// 104 void getAnalysisUsage(AnalysisUsage &AU) const override { 105 AU.addRequired<LoopInfo>(); 106 AU.addPreserved<LoopInfo>(); 107 AU.addRequiredID(LoopSimplifyID); 108 AU.addPreservedID(LoopSimplifyID); 109 AU.addRequiredID(LCSSAID); 110 AU.addPreservedID(LCSSAID); 111 AU.addRequired<ScalarEvolution>(); 112 AU.addPreserved<ScalarEvolution>(); 113 AU.addRequired<TargetTransformInfo>(); 114 // FIXME: Loop unroll requires LCSSA. And LCSSA requires dom info. 115 // If loop unroll does not preserve dom info then LCSSA pass on next 116 // loop will receive invalid dom info. 117 // For now, recreate dom info, if loop is unrolled. 118 AU.addPreserved<DominatorTreeWrapperPass>(); 119 } 120 121 // Fill in the UnrollingPreferences parameter with values from the 122 // TargetTransformationInfo. 123 void getUnrollingPreferences(Loop *L, const TargetTransformInfo &TTI, 124 TargetTransformInfo::UnrollingPreferences &UP) { 125 UP.Threshold = CurrentThreshold; 126 UP.OptSizeThreshold = OptSizeUnrollThreshold; 127 UP.PartialThreshold = CurrentThreshold; 128 UP.PartialOptSizeThreshold = OptSizeUnrollThreshold; 129 UP.Count = CurrentCount; 130 UP.MaxCount = UINT_MAX; 131 UP.Partial = CurrentAllowPartial; 132 UP.Runtime = CurrentRuntime; 133 TTI.getUnrollingPreferences(L, UP); 134 } 135 136 // Select and return an unroll count based on parameters from 137 // user, unroll preferences, unroll pragmas, or a heuristic. 138 // SetExplicitly is set to true if the unroll count is is set by 139 // the user or a pragma rather than selected heuristically. 140 unsigned 141 selectUnrollCount(const Loop *L, unsigned TripCount, bool HasEnablePragma, 142 unsigned PragmaCount, 143 const TargetTransformInfo::UnrollingPreferences &UP, 144 bool &SetExplicitly); 145 146 147 // Select threshold values used to limit unrolling based on a 148 // total unrolled size. Parameters Threshold and PartialThreshold 149 // are set to the maximum unrolled size for fully and partially 150 // unrolled loops respectively. 151 void selectThresholds(const Loop *L, bool HasPragma, 152 const TargetTransformInfo::UnrollingPreferences &UP, 153 unsigned &Threshold, unsigned &PartialThreshold) { 154 // Determine the current unrolling threshold. While this is 155 // normally set from UnrollThreshold, it is overridden to a 156 // smaller value if the current function is marked as 157 // optimize-for-size, and the unroll threshold was not user 158 // specified. 159 Threshold = UserThreshold ? CurrentThreshold : UP.Threshold; 160 PartialThreshold = UserThreshold ? CurrentThreshold : UP.PartialThreshold; 161 if (!UserThreshold && 162 L->getHeader()->getParent()->getAttributes(). 163 hasAttribute(AttributeSet::FunctionIndex, 164 Attribute::OptimizeForSize)) { 165 Threshold = UP.OptSizeThreshold; 166 PartialThreshold = UP.PartialOptSizeThreshold; 167 } 168 if (HasPragma) { 169 // If the loop has an unrolling pragma, we want to be more 170 // aggressive with unrolling limits. Set thresholds to at 171 // least the PragmaTheshold value which is larger than the 172 // default limits. 173 if (Threshold != NoThreshold) 174 Threshold = std::max<unsigned>(Threshold, PragmaUnrollThreshold); 175 if (PartialThreshold != NoThreshold) 176 PartialThreshold = 177 std::max<unsigned>(PartialThreshold, PragmaUnrollThreshold); 178 } 179 } 180 }; 181 } 182 183 char LoopUnroll::ID = 0; 184 INITIALIZE_PASS_BEGIN(LoopUnroll, "loop-unroll", "Unroll loops", false, false) 185 INITIALIZE_AG_DEPENDENCY(TargetTransformInfo) 186 INITIALIZE_PASS_DEPENDENCY(LoopInfo) 187 INITIALIZE_PASS_DEPENDENCY(LoopSimplify) 188 INITIALIZE_PASS_DEPENDENCY(LCSSA) 189 INITIALIZE_PASS_DEPENDENCY(ScalarEvolution) 190 INITIALIZE_PASS_END(LoopUnroll, "loop-unroll", "Unroll loops", false, false) 191 192 Pass *llvm::createLoopUnrollPass(int Threshold, int Count, int AllowPartial, 193 int Runtime) { 194 return new LoopUnroll(Threshold, Count, AllowPartial, Runtime); 195 } 196 197 Pass *llvm::createSimpleLoopUnrollPass() { 198 return llvm::createLoopUnrollPass(-1, -1, 0, 0); 199 } 200 201 /// ApproximateLoopSize - Approximate the size of the loop. 202 static unsigned ApproximateLoopSize(const Loop *L, unsigned &NumCalls, 203 bool &NotDuplicatable, 204 const TargetTransformInfo &TTI) { 205 CodeMetrics Metrics; 206 for (Loop::block_iterator I = L->block_begin(), E = L->block_end(); 207 I != E; ++I) 208 Metrics.analyzeBasicBlock(*I, TTI); 209 NumCalls = Metrics.NumInlineCandidates; 210 NotDuplicatable = Metrics.notDuplicatable; 211 212 unsigned LoopSize = Metrics.NumInsts; 213 214 // Don't allow an estimate of size zero. This would allows unrolling of loops 215 // with huge iteration counts, which is a compile time problem even if it's 216 // not a problem for code quality. 217 if (LoopSize == 0) LoopSize = 1; 218 219 return LoopSize; 220 } 221 222 // Returns the value associated with the given metadata node name (for 223 // example, "llvm.loop.unroll.count"). If no such named metadata node 224 // exists, then nullptr is returned. 225 static const ConstantInt *GetUnrollMetadataValue(const Loop *L, 226 StringRef Name) { 227 MDNode *LoopID = L->getLoopID(); 228 if (!LoopID) return nullptr; 229 230 // First operand should refer to the loop id itself. 231 assert(LoopID->getNumOperands() > 0 && "requires at least one operand"); 232 assert(LoopID->getOperand(0) == LoopID && "invalid loop id"); 233 234 for (unsigned i = 1, e = LoopID->getNumOperands(); i < e; ++i) { 235 const MDNode *MD = dyn_cast<MDNode>(LoopID->getOperand(i)); 236 if (!MD) continue; 237 238 const MDString *S = dyn_cast<MDString>(MD->getOperand(0)); 239 if (!S) continue; 240 241 if (Name.equals(S->getString())) { 242 assert(MD->getNumOperands() == 2 && 243 "Unroll hint metadata should have two operands."); 244 return cast<ConstantInt>(MD->getOperand(1)); 245 } 246 } 247 return nullptr; 248 } 249 250 // Returns true if the loop has an unroll(enable) pragma. 251 static bool HasUnrollEnablePragma(const Loop *L) { 252 const ConstantInt *EnableValue = 253 GetUnrollMetadataValue(L, "llvm.loop.unroll.enable"); 254 return (EnableValue && EnableValue->getZExtValue()); 255 } 256 257 // Returns true if the loop has an unroll(disable) pragma. 258 static bool HasUnrollDisablePragma(const Loop *L) { 259 const ConstantInt *EnableValue = 260 GetUnrollMetadataValue(L, "llvm.loop.unroll.enable"); 261 return (EnableValue && !EnableValue->getZExtValue()); 262 } 263 264 // If loop has an unroll_count pragma return the (necessarily 265 // positive) value from the pragma. Otherwise return 0. 266 static unsigned UnrollCountPragmaValue(const Loop *L) { 267 const ConstantInt *CountValue = 268 GetUnrollMetadataValue(L, "llvm.loop.unroll.count"); 269 if (CountValue) { 270 unsigned Count = CountValue->getZExtValue(); 271 assert(Count >= 1 && "Unroll count must be positive."); 272 return Count; 273 } 274 return 0; 275 } 276 277 // Remove existing unroll metadata and add unroll disable metadata to 278 // indicate the loop has already been unrolled. This prevents a loop 279 // from being unrolled more than is directed by a pragma if the loop 280 // unrolling pass is run more than once (which it generally is). 281 static void SetLoopAlreadyUnrolled(Loop *L) { 282 MDNode *LoopID = L->getLoopID(); 283 if (!LoopID) return; 284 285 // First remove any existing loop unrolling metadata. 286 SmallVector<Value *, 4> Vals; 287 // Reserve first location for self reference to the LoopID metadata node. 288 Vals.push_back(nullptr); 289 for (unsigned i = 1, ie = LoopID->getNumOperands(); i < ie; ++i) { 290 bool IsUnrollMetadata = false; 291 MDNode *MD = dyn_cast<MDNode>(LoopID->getOperand(i)); 292 if (MD) { 293 const MDString *S = dyn_cast<MDString>(MD->getOperand(0)); 294 IsUnrollMetadata = S && S->getString().startswith("llvm.loop.unroll."); 295 } 296 if (!IsUnrollMetadata) Vals.push_back(LoopID->getOperand(i)); 297 } 298 299 // Add unroll(disable) metadata to disable future unrolling. 300 LLVMContext &Context = L->getHeader()->getContext(); 301 SmallVector<Value *, 2> DisableOperands; 302 DisableOperands.push_back(MDString::get(Context, "llvm.loop.unroll.enable")); 303 DisableOperands.push_back(ConstantInt::get(Type::getInt1Ty(Context), 0)); 304 MDNode *DisableNode = MDNode::get(Context, DisableOperands); 305 Vals.push_back(DisableNode); 306 307 MDNode *NewLoopID = MDNode::get(Context, Vals); 308 // Set operand 0 to refer to the loop id itself. 309 NewLoopID->replaceOperandWith(0, NewLoopID); 310 L->setLoopID(NewLoopID); 311 LoopID->replaceAllUsesWith(NewLoopID); 312 } 313 314 unsigned LoopUnroll::selectUnrollCount( 315 const Loop *L, unsigned TripCount, bool HasEnablePragma, 316 unsigned PragmaCount, const TargetTransformInfo::UnrollingPreferences &UP, 317 bool &SetExplicitly) { 318 SetExplicitly = true; 319 320 // User-specified count (either as a command-line option or 321 // constructor parameter) has highest precedence. 322 unsigned Count = UserCount ? CurrentCount : 0; 323 324 // If there is no user-specified count, unroll pragmas have the next 325 // highest precendence. 326 if (Count == 0) { 327 if (PragmaCount) { 328 Count = PragmaCount; 329 } else if (HasEnablePragma) { 330 // unroll(enable) pragma without an unroll_count pragma 331 // indicates to unroll loop fully. 332 Count = TripCount; 333 } 334 } 335 336 if (Count == 0) 337 Count = UP.Count; 338 339 if (Count == 0) { 340 SetExplicitly = false; 341 if (TripCount == 0) 342 // Runtime trip count. 343 Count = UnrollRuntimeCount; 344 else 345 // Conservative heuristic: if we know the trip count, see if we can 346 // completely unroll (subject to the threshold, checked below); otherwise 347 // try to find greatest modulo of the trip count which is still under 348 // threshold value. 349 Count = TripCount; 350 } 351 if (TripCount && Count > TripCount) 352 return TripCount; 353 return Count; 354 } 355 356 bool LoopUnroll::runOnLoop(Loop *L, LPPassManager &LPM) { 357 if (skipOptnoneFunction(L)) 358 return false; 359 360 LoopInfo *LI = &getAnalysis<LoopInfo>(); 361 ScalarEvolution *SE = &getAnalysis<ScalarEvolution>(); 362 const TargetTransformInfo &TTI = getAnalysis<TargetTransformInfo>(); 363 364 BasicBlock *Header = L->getHeader(); 365 DEBUG(dbgs() << "Loop Unroll: F[" << Header->getParent()->getName() 366 << "] Loop %" << Header->getName() << "\n"); 367 368 if (HasUnrollDisablePragma(L)) { 369 return false; 370 } 371 bool HasEnablePragma = HasUnrollEnablePragma(L); 372 unsigned PragmaCount = UnrollCountPragmaValue(L); 373 bool HasPragma = HasEnablePragma || PragmaCount > 0; 374 375 TargetTransformInfo::UnrollingPreferences UP; 376 getUnrollingPreferences(L, TTI, UP); 377 378 // Find trip count and trip multiple if count is not available 379 unsigned TripCount = 0; 380 unsigned TripMultiple = 1; 381 // Find "latch trip count". UnrollLoop assumes that control cannot exit 382 // via the loop latch on any iteration prior to TripCount. The loop may exit 383 // early via an earlier branch. 384 BasicBlock *LatchBlock = L->getLoopLatch(); 385 if (LatchBlock) { 386 TripCount = SE->getSmallConstantTripCount(L, LatchBlock); 387 TripMultiple = SE->getSmallConstantTripMultiple(L, LatchBlock); 388 } 389 390 // Select an initial unroll count. This may be reduced later based 391 // on size thresholds. 392 bool CountSetExplicitly; 393 unsigned Count = selectUnrollCount(L, TripCount, HasEnablePragma, PragmaCount, 394 UP, CountSetExplicitly); 395 396 unsigned NumInlineCandidates; 397 bool notDuplicatable; 398 unsigned LoopSize = 399 ApproximateLoopSize(L, NumInlineCandidates, notDuplicatable, TTI); 400 DEBUG(dbgs() << " Loop Size = " << LoopSize << "\n"); 401 uint64_t UnrolledSize = (uint64_t)LoopSize * Count; 402 if (notDuplicatable) { 403 DEBUG(dbgs() << " Not unrolling loop which contains non-duplicatable" 404 << " instructions.\n"); 405 return false; 406 } 407 if (NumInlineCandidates != 0) { 408 DEBUG(dbgs() << " Not unrolling loop with inlinable calls.\n"); 409 return false; 410 } 411 412 unsigned Threshold, PartialThreshold; 413 selectThresholds(L, HasPragma, UP, Threshold, PartialThreshold); 414 415 // Given Count, TripCount and thresholds determine the type of 416 // unrolling which is to be performed. 417 enum { Full = 0, Partial = 1, Runtime = 2 }; 418 int Unrolling; 419 if (TripCount && Count == TripCount) { 420 if (Threshold != NoThreshold && UnrolledSize > Threshold) { 421 DEBUG(dbgs() << " Too large to fully unroll with count: " << Count 422 << " because size: " << UnrolledSize << ">" << Threshold 423 << "\n"); 424 Unrolling = Partial; 425 } else { 426 Unrolling = Full; 427 } 428 } else if (TripCount && Count < TripCount) { 429 Unrolling = Partial; 430 } else { 431 Unrolling = Runtime; 432 } 433 434 // Reduce count based on the type of unrolling and the threshold values. 435 unsigned OriginalCount = Count; 436 bool AllowRuntime = UserRuntime ? CurrentRuntime : UP.Runtime; 437 if (Unrolling == Partial) { 438 bool AllowPartial = UserAllowPartial ? CurrentAllowPartial : UP.Partial; 439 if (!AllowPartial && !CountSetExplicitly) { 440 DEBUG(dbgs() << " will not try to unroll partially because " 441 << "-unroll-allow-partial not given\n"); 442 return false; 443 } 444 if (PartialThreshold != NoThreshold && UnrolledSize > PartialThreshold) { 445 // Reduce unroll count to be modulo of TripCount for partial unrolling. 446 Count = PartialThreshold / LoopSize; 447 while (Count != 0 && TripCount % Count != 0) 448 Count--; 449 } 450 } else if (Unrolling == Runtime) { 451 if (!AllowRuntime && !CountSetExplicitly) { 452 DEBUG(dbgs() << " will not try to unroll loop with runtime trip count " 453 << "-unroll-runtime not given\n"); 454 return false; 455 } 456 // Reduce unroll count to be the largest power-of-two factor of 457 // the original count which satisfies the threshold limit. 458 while (Count != 0 && UnrolledSize > PartialThreshold) { 459 Count >>= 1; 460 UnrolledSize = LoopSize * Count; 461 } 462 if (Count > UP.MaxCount) 463 Count = UP.MaxCount; 464 DEBUG(dbgs() << " partially unrolling with count: " << Count << "\n"); 465 } 466 467 if (HasPragma) { 468 // Mark loop as unrolled to prevent unrolling beyond that 469 // requested by the pragma. 470 SetLoopAlreadyUnrolled(L); 471 472 // Emit optimization remarks if we are unable to unroll the loop 473 // as directed by a pragma. 474 DebugLoc LoopLoc = L->getStartLoc(); 475 Function *F = Header->getParent(); 476 LLVMContext &Ctx = F->getContext(); 477 if (HasEnablePragma && PragmaCount == 0) { 478 if (TripCount && Count != TripCount) { 479 emitOptimizationRemarkMissed( 480 Ctx, DEBUG_TYPE, *F, LoopLoc, 481 "Unable to fully unroll loop as directed by unroll(enable) pragma " 482 "because unrolled size is too large."); 483 } else if (!TripCount) { 484 emitOptimizationRemarkMissed( 485 Ctx, DEBUG_TYPE, *F, LoopLoc, 486 "Unable to fully unroll loop as directed by unroll(enable) pragma " 487 "because loop has a runtime trip count."); 488 } 489 } else if (PragmaCount > 0 && Count != OriginalCount) { 490 emitOptimizationRemarkMissed( 491 Ctx, DEBUG_TYPE, *F, LoopLoc, 492 "Unable to unroll loop the number of times directed by " 493 "unroll_count pragma because unrolled size is too large."); 494 } 495 } 496 497 if (Unrolling != Full && Count < 2) { 498 // Partial unrolling by 1 is a nop. For full unrolling, a factor 499 // of 1 makes sense because loop control can be eliminated. 500 return false; 501 } 502 503 // Unroll the loop. 504 if (!UnrollLoop(L, Count, TripCount, AllowRuntime, TripMultiple, LI, this, &LPM)) 505 return false; 506 507 return true; 508 } 509