1 //===-- StraightLineStrengthReduce.cpp - ------------------------*- C++ -*-===// 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 file implements straight-line strength reduction (SLSR). Unlike loop 11 // strength reduction, this algorithm is designed to reduce arithmetic 12 // redundancy in straight-line code instead of loops. It has proven to be 13 // effective in simplifying arithmetic statements derived from an unrolled loop. 14 // It can also simplify the logic of SeparateConstOffsetFromGEP. 15 // 16 // There are many optimizations we can perform in the domain of SLSR. This file 17 // for now contains only an initial step. Specifically, we look for strength 18 // reduction candidates in the following forms: 19 // 20 // Form 1: B + i * S 21 // Form 2: (B + i) * S 22 // Form 3: &B[i * S] 23 // 24 // where S is an integer variable, and i is a constant integer. If we found two 25 // candidates S1 and S2 in the same form and S1 dominates S2, we may rewrite S2 26 // in a simpler way with respect to S1. For example, 27 // 28 // S1: X = B + i * S 29 // S2: Y = B + i' * S => X + (i' - i) * S 30 // 31 // S1: X = (B + i) * S 32 // S2: Y = (B + i') * S => X + (i' - i) * S 33 // 34 // S1: X = &B[i * S] 35 // S2: Y = &B[i' * S] => &X[(i' - i) * S] 36 // 37 // Note: (i' - i) * S is folded to the extent possible. 38 // 39 // This rewriting is in general a good idea. The code patterns we focus on 40 // usually come from loop unrolling, so (i' - i) * S is likely the same 41 // across iterations and can be reused. When that happens, the optimized form 42 // takes only one add starting from the second iteration. 43 // 44 // When such rewriting is possible, we call S1 a "basis" of S2. When S2 has 45 // multiple bases, we choose to rewrite S2 with respect to its "immediate" 46 // basis, the basis that is the closest ancestor in the dominator tree. 47 // 48 // TODO: 49 // 50 // - Floating point arithmetics when fast math is enabled. 51 // 52 // - SLSR may decrease ILP at the architecture level. Targets that are very 53 // sensitive to ILP may want to disable it. Having SLSR to consider ILP is 54 // left as future work. 55 // 56 // - When (i' - i) is constant but i and i' are not, we could still perform 57 // SLSR. 58 #include <vector> 59 60 #include "llvm/Analysis/ScalarEvolution.h" 61 #include "llvm/Analysis/TargetTransformInfo.h" 62 #include "llvm/Analysis/ValueTracking.h" 63 #include "llvm/IR/DataLayout.h" 64 #include "llvm/IR/Dominators.h" 65 #include "llvm/IR/IRBuilder.h" 66 #include "llvm/IR/Module.h" 67 #include "llvm/IR/PatternMatch.h" 68 #include "llvm/Support/raw_ostream.h" 69 #include "llvm/Transforms/Scalar.h" 70 #include "llvm/Transforms/Utils/Local.h" 71 72 using namespace llvm; 73 using namespace PatternMatch; 74 75 namespace { 76 77 static const unsigned UnknownAddressSpace = ~0u; 78 79 class StraightLineStrengthReduce : public FunctionPass { 80 public: 81 // SLSR candidate. Such a candidate must be in one of the forms described in 82 // the header comments. 83 struct Candidate : public ilist_node<Candidate> { 84 enum Kind { 85 Invalid, // reserved for the default constructor 86 Add, // B + i * S 87 Mul, // (B + i) * S 88 GEP, // &B[..][i * S][..] 89 }; 90 91 Candidate() 92 : CandidateKind(Invalid), Base(nullptr), Index(nullptr), 93 Stride(nullptr), Ins(nullptr), Basis(nullptr) {} 94 Candidate(Kind CT, const SCEV *B, ConstantInt *Idx, Value *S, 95 Instruction *I) 96 : CandidateKind(CT), Base(B), Index(Idx), Stride(S), Ins(I), 97 Basis(nullptr) {} 98 Kind CandidateKind; 99 const SCEV *Base; 100 // Note that Index and Stride of a GEP candidate do not necessarily have the 101 // same integer type. In that case, during rewriting, Stride will be 102 // sign-extended or truncated to Index's type. 103 ConstantInt *Index; 104 Value *Stride; 105 // The instruction this candidate corresponds to. It helps us to rewrite a 106 // candidate with respect to its immediate basis. Note that one instruction 107 // can correspond to multiple candidates depending on how you associate the 108 // expression. For instance, 109 // 110 // (a + 1) * (b + 2) 111 // 112 // can be treated as 113 // 114 // <Base: a, Index: 1, Stride: b + 2> 115 // 116 // or 117 // 118 // <Base: b, Index: 2, Stride: a + 1> 119 Instruction *Ins; 120 // Points to the immediate basis of this candidate, or nullptr if we cannot 121 // find any basis for this candidate. 122 Candidate *Basis; 123 }; 124 125 static char ID; 126 127 StraightLineStrengthReduce() 128 : FunctionPass(ID), DL(nullptr), DT(nullptr), TTI(nullptr) { 129 initializeStraightLineStrengthReducePass(*PassRegistry::getPassRegistry()); 130 } 131 132 void getAnalysisUsage(AnalysisUsage &AU) const override { 133 AU.addRequired<DominatorTreeWrapperPass>(); 134 AU.addRequired<ScalarEvolutionWrapperPass>(); 135 AU.addRequired<TargetTransformInfoWrapperPass>(); 136 // We do not modify the shape of the CFG. 137 AU.setPreservesCFG(); 138 } 139 140 bool doInitialization(Module &M) override { 141 DL = &M.getDataLayout(); 142 return false; 143 } 144 145 bool runOnFunction(Function &F) override; 146 147 private: 148 // Returns true if Basis is a basis for C, i.e., Basis dominates C and they 149 // share the same base and stride. 150 bool isBasisFor(const Candidate &Basis, const Candidate &C); 151 // Returns whether the candidate can be folded into an addressing mode. 152 bool isFoldable(const Candidate &C, TargetTransformInfo *TTI, 153 const DataLayout *DL); 154 // Returns true if C is already in a simplest form and not worth being 155 // rewritten. 156 bool isSimplestForm(const Candidate &C); 157 // Checks whether I is in a candidate form. If so, adds all the matching forms 158 // to Candidates, and tries to find the immediate basis for each of them. 159 void allocateCandidatesAndFindBasis(Instruction *I); 160 // Allocate candidates and find bases for Add instructions. 161 void allocateCandidatesAndFindBasisForAdd(Instruction *I); 162 // Given I = LHS + RHS, factors RHS into i * S and makes (LHS + i * S) a 163 // candidate. 164 void allocateCandidatesAndFindBasisForAdd(Value *LHS, Value *RHS, 165 Instruction *I); 166 // Allocate candidates and find bases for Mul instructions. 167 void allocateCandidatesAndFindBasisForMul(Instruction *I); 168 // Splits LHS into Base + Index and, if succeeds, calls 169 // allocateCandidatesAndFindBasis. 170 void allocateCandidatesAndFindBasisForMul(Value *LHS, Value *RHS, 171 Instruction *I); 172 // Allocate candidates and find bases for GetElementPtr instructions. 173 void allocateCandidatesAndFindBasisForGEP(GetElementPtrInst *GEP); 174 // A helper function that scales Idx with ElementSize before invoking 175 // allocateCandidatesAndFindBasis. 176 void allocateCandidatesAndFindBasisForGEP(const SCEV *B, ConstantInt *Idx, 177 Value *S, uint64_t ElementSize, 178 Instruction *I); 179 // Adds the given form <CT, B, Idx, S> to Candidates, and finds its immediate 180 // basis. 181 void allocateCandidatesAndFindBasis(Candidate::Kind CT, const SCEV *B, 182 ConstantInt *Idx, Value *S, 183 Instruction *I); 184 // Rewrites candidate C with respect to Basis. 185 void rewriteCandidateWithBasis(const Candidate &C, const Candidate &Basis); 186 // A helper function that factors ArrayIdx to a product of a stride and a 187 // constant index, and invokes allocateCandidatesAndFindBasis with the 188 // factorings. 189 void factorArrayIndex(Value *ArrayIdx, const SCEV *Base, uint64_t ElementSize, 190 GetElementPtrInst *GEP); 191 // Emit code that computes the "bump" from Basis to C. If the candidate is a 192 // GEP and the bump is not divisible by the element size of the GEP, this 193 // function sets the BumpWithUglyGEP flag to notify its caller to bump the 194 // basis using an ugly GEP. 195 static Value *emitBump(const Candidate &Basis, const Candidate &C, 196 IRBuilder<> &Builder, const DataLayout *DL, 197 bool &BumpWithUglyGEP); 198 199 const DataLayout *DL; 200 DominatorTree *DT; 201 ScalarEvolution *SE; 202 TargetTransformInfo *TTI; 203 ilist<Candidate> Candidates; 204 // Temporarily holds all instructions that are unlinked (but not deleted) by 205 // rewriteCandidateWithBasis. These instructions will be actually removed 206 // after all rewriting finishes. 207 std::vector<Instruction *> UnlinkedInstructions; 208 }; 209 } // anonymous namespace 210 211 char StraightLineStrengthReduce::ID = 0; 212 INITIALIZE_PASS_BEGIN(StraightLineStrengthReduce, "slsr", 213 "Straight line strength reduction", false, false) 214 INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass) 215 INITIALIZE_PASS_DEPENDENCY(ScalarEvolutionWrapperPass) 216 INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass) 217 INITIALIZE_PASS_END(StraightLineStrengthReduce, "slsr", 218 "Straight line strength reduction", false, false) 219 220 FunctionPass *llvm::createStraightLineStrengthReducePass() { 221 return new StraightLineStrengthReduce(); 222 } 223 224 bool StraightLineStrengthReduce::isBasisFor(const Candidate &Basis, 225 const Candidate &C) { 226 return (Basis.Ins != C.Ins && // skip the same instruction 227 // They must have the same type too. Basis.Base == C.Base doesn't 228 // guarantee their types are the same (PR23975). 229 Basis.Ins->getType() == C.Ins->getType() && 230 // Basis must dominate C in order to rewrite C with respect to Basis. 231 DT->dominates(Basis.Ins->getParent(), C.Ins->getParent()) && 232 // They share the same base, stride, and candidate kind. 233 Basis.Base == C.Base && Basis.Stride == C.Stride && 234 Basis.CandidateKind == C.CandidateKind); 235 } 236 237 static bool isGEPFoldable(GetElementPtrInst *GEP, 238 const TargetTransformInfo *TTI) { 239 SmallVector<const Value*, 4> Indices; 240 for (auto I = GEP->idx_begin(); I != GEP->idx_end(); ++I) 241 Indices.push_back(*I); 242 return TTI->getGEPCost(GEP->getSourceElementType(), GEP->getPointerOperand(), 243 Indices) == TargetTransformInfo::TCC_Free; 244 } 245 246 // Returns whether (Base + Index * Stride) can be folded to an addressing mode. 247 static bool isAddFoldable(const SCEV *Base, ConstantInt *Index, Value *Stride, 248 TargetTransformInfo *TTI) { 249 // Index->getSExtValue() may crash if Index is wider than 64-bit. 250 return Index->getBitWidth() <= 64 && 251 TTI->isLegalAddressingMode(Base->getType(), nullptr, 0, true, 252 Index->getSExtValue(), UnknownAddressSpace); 253 } 254 255 bool StraightLineStrengthReduce::isFoldable(const Candidate &C, 256 TargetTransformInfo *TTI, 257 const DataLayout *DL) { 258 if (C.CandidateKind == Candidate::Add) 259 return isAddFoldable(C.Base, C.Index, C.Stride, TTI); 260 if (C.CandidateKind == Candidate::GEP) 261 return isGEPFoldable(cast<GetElementPtrInst>(C.Ins), TTI); 262 return false; 263 } 264 265 // Returns true if GEP has zero or one non-zero index. 266 static bool hasOnlyOneNonZeroIndex(GetElementPtrInst *GEP) { 267 unsigned NumNonZeroIndices = 0; 268 for (auto I = GEP->idx_begin(); I != GEP->idx_end(); ++I) { 269 ConstantInt *ConstIdx = dyn_cast<ConstantInt>(*I); 270 if (ConstIdx == nullptr || !ConstIdx->isZero()) 271 ++NumNonZeroIndices; 272 } 273 return NumNonZeroIndices <= 1; 274 } 275 276 bool StraightLineStrengthReduce::isSimplestForm(const Candidate &C) { 277 if (C.CandidateKind == Candidate::Add) { 278 // B + 1 * S or B + (-1) * S 279 return C.Index->isOne() || C.Index->isMinusOne(); 280 } 281 if (C.CandidateKind == Candidate::Mul) { 282 // (B + 0) * S 283 return C.Index->isZero(); 284 } 285 if (C.CandidateKind == Candidate::GEP) { 286 // (char*)B + S or (char*)B - S 287 return ((C.Index->isOne() || C.Index->isMinusOne()) && 288 hasOnlyOneNonZeroIndex(cast<GetElementPtrInst>(C.Ins))); 289 } 290 return false; 291 } 292 293 // TODO: We currently implement an algorithm whose time complexity is linear in 294 // the number of existing candidates. However, we could do better by using 295 // ScopedHashTable. Specifically, while traversing the dominator tree, we could 296 // maintain all the candidates that dominate the basic block being traversed in 297 // a ScopedHashTable. This hash table is indexed by the base and the stride of 298 // a candidate. Therefore, finding the immediate basis of a candidate boils down 299 // to one hash-table look up. 300 void StraightLineStrengthReduce::allocateCandidatesAndFindBasis( 301 Candidate::Kind CT, const SCEV *B, ConstantInt *Idx, Value *S, 302 Instruction *I) { 303 Candidate C(CT, B, Idx, S, I); 304 // SLSR can complicate an instruction in two cases: 305 // 306 // 1. If we can fold I into an addressing mode, computing I is likely free or 307 // takes only one instruction. 308 // 309 // 2. I is already in a simplest form. For example, when 310 // X = B + 8 * S 311 // Y = B + S, 312 // rewriting Y to X - 7 * S is probably a bad idea. 313 // 314 // In the above cases, we still add I to the candidate list so that I can be 315 // the basis of other candidates, but we leave I's basis blank so that I 316 // won't be rewritten. 317 if (!isFoldable(C, TTI, DL) && !isSimplestForm(C)) { 318 // Try to compute the immediate basis of C. 319 unsigned NumIterations = 0; 320 // Limit the scan radius to avoid running in quadratice time. 321 static const unsigned MaxNumIterations = 50; 322 for (auto Basis = Candidates.rbegin(); 323 Basis != Candidates.rend() && NumIterations < MaxNumIterations; 324 ++Basis, ++NumIterations) { 325 if (isBasisFor(*Basis, C)) { 326 C.Basis = &(*Basis); 327 break; 328 } 329 } 330 } 331 // Regardless of whether we find a basis for C, we need to push C to the 332 // candidate list so that it can be the basis of other candidates. 333 Candidates.push_back(C); 334 } 335 336 void StraightLineStrengthReduce::allocateCandidatesAndFindBasis( 337 Instruction *I) { 338 switch (I->getOpcode()) { 339 case Instruction::Add: 340 allocateCandidatesAndFindBasisForAdd(I); 341 break; 342 case Instruction::Mul: 343 allocateCandidatesAndFindBasisForMul(I); 344 break; 345 case Instruction::GetElementPtr: 346 allocateCandidatesAndFindBasisForGEP(cast<GetElementPtrInst>(I)); 347 break; 348 } 349 } 350 351 void StraightLineStrengthReduce::allocateCandidatesAndFindBasisForAdd( 352 Instruction *I) { 353 // Try matching B + i * S. 354 if (!isa<IntegerType>(I->getType())) 355 return; 356 357 assert(I->getNumOperands() == 2 && "isn't I an add?"); 358 Value *LHS = I->getOperand(0), *RHS = I->getOperand(1); 359 allocateCandidatesAndFindBasisForAdd(LHS, RHS, I); 360 if (LHS != RHS) 361 allocateCandidatesAndFindBasisForAdd(RHS, LHS, I); 362 } 363 364 void StraightLineStrengthReduce::allocateCandidatesAndFindBasisForAdd( 365 Value *LHS, Value *RHS, Instruction *I) { 366 Value *S = nullptr; 367 ConstantInt *Idx = nullptr; 368 if (match(RHS, m_Mul(m_Value(S), m_ConstantInt(Idx)))) { 369 // I = LHS + RHS = LHS + Idx * S 370 allocateCandidatesAndFindBasis(Candidate::Add, SE->getSCEV(LHS), Idx, S, I); 371 } else if (match(RHS, m_Shl(m_Value(S), m_ConstantInt(Idx)))) { 372 // I = LHS + RHS = LHS + (S << Idx) = LHS + S * (1 << Idx) 373 APInt One(Idx->getBitWidth(), 1); 374 Idx = ConstantInt::get(Idx->getContext(), One << Idx->getValue()); 375 allocateCandidatesAndFindBasis(Candidate::Add, SE->getSCEV(LHS), Idx, S, I); 376 } else { 377 // At least, I = LHS + 1 * RHS 378 ConstantInt *One = ConstantInt::get(cast<IntegerType>(I->getType()), 1); 379 allocateCandidatesAndFindBasis(Candidate::Add, SE->getSCEV(LHS), One, RHS, 380 I); 381 } 382 } 383 384 // Returns true if A matches B + C where C is constant. 385 static bool matchesAdd(Value *A, Value *&B, ConstantInt *&C) { 386 return (match(A, m_Add(m_Value(B), m_ConstantInt(C))) || 387 match(A, m_Add(m_ConstantInt(C), m_Value(B)))); 388 } 389 390 // Returns true if A matches B | C where C is constant. 391 static bool matchesOr(Value *A, Value *&B, ConstantInt *&C) { 392 return (match(A, m_Or(m_Value(B), m_ConstantInt(C))) || 393 match(A, m_Or(m_ConstantInt(C), m_Value(B)))); 394 } 395 396 void StraightLineStrengthReduce::allocateCandidatesAndFindBasisForMul( 397 Value *LHS, Value *RHS, Instruction *I) { 398 Value *B = nullptr; 399 ConstantInt *Idx = nullptr; 400 if (matchesAdd(LHS, B, Idx)) { 401 // If LHS is in the form of "Base + Index", then I is in the form of 402 // "(Base + Index) * RHS". 403 allocateCandidatesAndFindBasis(Candidate::Mul, SE->getSCEV(B), Idx, RHS, I); 404 } else if (matchesOr(LHS, B, Idx) && haveNoCommonBitsSet(B, Idx, *DL)) { 405 // If LHS is in the form of "Base | Index" and Base and Index have no common 406 // bits set, then 407 // Base | Index = Base + Index 408 // and I is thus in the form of "(Base + Index) * RHS". 409 allocateCandidatesAndFindBasis(Candidate::Mul, SE->getSCEV(B), Idx, RHS, I); 410 } else { 411 // Otherwise, at least try the form (LHS + 0) * RHS. 412 ConstantInt *Zero = ConstantInt::get(cast<IntegerType>(I->getType()), 0); 413 allocateCandidatesAndFindBasis(Candidate::Mul, SE->getSCEV(LHS), Zero, RHS, 414 I); 415 } 416 } 417 418 void StraightLineStrengthReduce::allocateCandidatesAndFindBasisForMul( 419 Instruction *I) { 420 // Try matching (B + i) * S. 421 // TODO: we could extend SLSR to float and vector types. 422 if (!isa<IntegerType>(I->getType())) 423 return; 424 425 assert(I->getNumOperands() == 2 && "isn't I a mul?"); 426 Value *LHS = I->getOperand(0), *RHS = I->getOperand(1); 427 allocateCandidatesAndFindBasisForMul(LHS, RHS, I); 428 if (LHS != RHS) { 429 // Symmetrically, try to split RHS to Base + Index. 430 allocateCandidatesAndFindBasisForMul(RHS, LHS, I); 431 } 432 } 433 434 void StraightLineStrengthReduce::allocateCandidatesAndFindBasisForGEP( 435 const SCEV *B, ConstantInt *Idx, Value *S, uint64_t ElementSize, 436 Instruction *I) { 437 // I = B + sext(Idx *nsw S) * ElementSize 438 // = B + (sext(Idx) * sext(S)) * ElementSize 439 // = B + (sext(Idx) * ElementSize) * sext(S) 440 // Casting to IntegerType is safe because we skipped vector GEPs. 441 IntegerType *IntPtrTy = cast<IntegerType>(DL->getIntPtrType(I->getType())); 442 ConstantInt *ScaledIdx = ConstantInt::get( 443 IntPtrTy, Idx->getSExtValue() * (int64_t)ElementSize, true); 444 allocateCandidatesAndFindBasis(Candidate::GEP, B, ScaledIdx, S, I); 445 } 446 447 void StraightLineStrengthReduce::factorArrayIndex(Value *ArrayIdx, 448 const SCEV *Base, 449 uint64_t ElementSize, 450 GetElementPtrInst *GEP) { 451 // At least, ArrayIdx = ArrayIdx *nsw 1. 452 allocateCandidatesAndFindBasisForGEP( 453 Base, ConstantInt::get(cast<IntegerType>(ArrayIdx->getType()), 1), 454 ArrayIdx, ElementSize, GEP); 455 Value *LHS = nullptr; 456 ConstantInt *RHS = nullptr; 457 // One alternative is matching the SCEV of ArrayIdx instead of ArrayIdx 458 // itself. This would allow us to handle the shl case for free. However, 459 // matching SCEVs has two issues: 460 // 461 // 1. this would complicate rewriting because the rewriting procedure 462 // would have to translate SCEVs back to IR instructions. This translation 463 // is difficult when LHS is further evaluated to a composite SCEV. 464 // 465 // 2. ScalarEvolution is designed to be control-flow oblivious. It tends 466 // to strip nsw/nuw flags which are critical for SLSR to trace into 467 // sext'ed multiplication. 468 if (match(ArrayIdx, m_NSWMul(m_Value(LHS), m_ConstantInt(RHS)))) { 469 // SLSR is currently unsafe if i * S may overflow. 470 // GEP = Base + sext(LHS *nsw RHS) * ElementSize 471 allocateCandidatesAndFindBasisForGEP(Base, RHS, LHS, ElementSize, GEP); 472 } else if (match(ArrayIdx, m_NSWShl(m_Value(LHS), m_ConstantInt(RHS)))) { 473 // GEP = Base + sext(LHS <<nsw RHS) * ElementSize 474 // = Base + sext(LHS *nsw (1 << RHS)) * ElementSize 475 APInt One(RHS->getBitWidth(), 1); 476 ConstantInt *PowerOf2 = 477 ConstantInt::get(RHS->getContext(), One << RHS->getValue()); 478 allocateCandidatesAndFindBasisForGEP(Base, PowerOf2, LHS, ElementSize, GEP); 479 } 480 } 481 482 void StraightLineStrengthReduce::allocateCandidatesAndFindBasisForGEP( 483 GetElementPtrInst *GEP) { 484 // TODO: handle vector GEPs 485 if (GEP->getType()->isVectorTy()) 486 return; 487 488 SmallVector<const SCEV *, 4> IndexExprs; 489 for (auto I = GEP->idx_begin(); I != GEP->idx_end(); ++I) 490 IndexExprs.push_back(SE->getSCEV(*I)); 491 492 gep_type_iterator GTI = gep_type_begin(GEP); 493 for (unsigned I = 1, E = GEP->getNumOperands(); I != E; ++I) { 494 if (!isa<SequentialType>(*GTI++)) 495 continue; 496 497 const SCEV *OrigIndexExpr = IndexExprs[I - 1]; 498 IndexExprs[I - 1] = SE->getZero(OrigIndexExpr->getType()); 499 500 // The base of this candidate is GEP's base plus the offsets of all 501 // indices except this current one. 502 const SCEV *BaseExpr = SE->getGEPExpr(GEP->getSourceElementType(), 503 SE->getSCEV(GEP->getPointerOperand()), 504 IndexExprs, GEP->isInBounds()); 505 Value *ArrayIdx = GEP->getOperand(I); 506 uint64_t ElementSize = DL->getTypeAllocSize(*GTI); 507 if (ArrayIdx->getType()->getIntegerBitWidth() <= 508 DL->getPointerSizeInBits(GEP->getAddressSpace())) { 509 // Skip factoring if ArrayIdx is wider than the pointer size, because 510 // ArrayIdx is implicitly truncated to the pointer size. 511 factorArrayIndex(ArrayIdx, BaseExpr, ElementSize, GEP); 512 } 513 // When ArrayIdx is the sext of a value, we try to factor that value as 514 // well. Handling this case is important because array indices are 515 // typically sign-extended to the pointer size. 516 Value *TruncatedArrayIdx = nullptr; 517 if (match(ArrayIdx, m_SExt(m_Value(TruncatedArrayIdx))) && 518 TruncatedArrayIdx->getType()->getIntegerBitWidth() <= 519 DL->getPointerSizeInBits(GEP->getAddressSpace())) { 520 // Skip factoring if TruncatedArrayIdx is wider than the pointer size, 521 // because TruncatedArrayIdx is implicitly truncated to the pointer size. 522 factorArrayIndex(TruncatedArrayIdx, BaseExpr, ElementSize, GEP); 523 } 524 525 IndexExprs[I - 1] = OrigIndexExpr; 526 } 527 } 528 529 // A helper function that unifies the bitwidth of A and B. 530 static void unifyBitWidth(APInt &A, APInt &B) { 531 if (A.getBitWidth() < B.getBitWidth()) 532 A = A.sext(B.getBitWidth()); 533 else if (A.getBitWidth() > B.getBitWidth()) 534 B = B.sext(A.getBitWidth()); 535 } 536 537 Value *StraightLineStrengthReduce::emitBump(const Candidate &Basis, 538 const Candidate &C, 539 IRBuilder<> &Builder, 540 const DataLayout *DL, 541 bool &BumpWithUglyGEP) { 542 APInt Idx = C.Index->getValue(), BasisIdx = Basis.Index->getValue(); 543 unifyBitWidth(Idx, BasisIdx); 544 APInt IndexOffset = Idx - BasisIdx; 545 546 BumpWithUglyGEP = false; 547 if (Basis.CandidateKind == Candidate::GEP) { 548 APInt ElementSize( 549 IndexOffset.getBitWidth(), 550 DL->getTypeAllocSize( 551 cast<GetElementPtrInst>(Basis.Ins)->getResultElementType())); 552 APInt Q, R; 553 APInt::sdivrem(IndexOffset, ElementSize, Q, R); 554 if (R == 0) 555 IndexOffset = Q; 556 else 557 BumpWithUglyGEP = true; 558 } 559 560 // Compute Bump = C - Basis = (i' - i) * S. 561 // Common case 1: if (i' - i) is 1, Bump = S. 562 if (IndexOffset == 1) 563 return C.Stride; 564 // Common case 2: if (i' - i) is -1, Bump = -S. 565 if (IndexOffset.isAllOnesValue()) 566 return Builder.CreateNeg(C.Stride); 567 568 // Otherwise, Bump = (i' - i) * sext/trunc(S). Note that (i' - i) and S may 569 // have different bit widths. 570 IntegerType *DeltaType = 571 IntegerType::get(Basis.Ins->getContext(), IndexOffset.getBitWidth()); 572 Value *ExtendedStride = Builder.CreateSExtOrTrunc(C.Stride, DeltaType); 573 if (IndexOffset.isPowerOf2()) { 574 // If (i' - i) is a power of 2, Bump = sext/trunc(S) << log(i' - i). 575 ConstantInt *Exponent = ConstantInt::get(DeltaType, IndexOffset.logBase2()); 576 return Builder.CreateShl(ExtendedStride, Exponent); 577 } 578 if ((-IndexOffset).isPowerOf2()) { 579 // If (i - i') is a power of 2, Bump = -sext/trunc(S) << log(i' - i). 580 ConstantInt *Exponent = 581 ConstantInt::get(DeltaType, (-IndexOffset).logBase2()); 582 return Builder.CreateNeg(Builder.CreateShl(ExtendedStride, Exponent)); 583 } 584 Constant *Delta = ConstantInt::get(DeltaType, IndexOffset); 585 return Builder.CreateMul(ExtendedStride, Delta); 586 } 587 588 void StraightLineStrengthReduce::rewriteCandidateWithBasis( 589 const Candidate &C, const Candidate &Basis) { 590 assert(C.CandidateKind == Basis.CandidateKind && C.Base == Basis.Base && 591 C.Stride == Basis.Stride); 592 // We run rewriteCandidateWithBasis on all candidates in a post-order, so the 593 // basis of a candidate cannot be unlinked before the candidate. 594 assert(Basis.Ins->getParent() != nullptr && "the basis is unlinked"); 595 596 // An instruction can correspond to multiple candidates. Therefore, instead of 597 // simply deleting an instruction when we rewrite it, we mark its parent as 598 // nullptr (i.e. unlink it) so that we can skip the candidates whose 599 // instruction is already rewritten. 600 if (!C.Ins->getParent()) 601 return; 602 603 IRBuilder<> Builder(C.Ins); 604 bool BumpWithUglyGEP; 605 Value *Bump = emitBump(Basis, C, Builder, DL, BumpWithUglyGEP); 606 Value *Reduced = nullptr; // equivalent to but weaker than C.Ins 607 switch (C.CandidateKind) { 608 case Candidate::Add: 609 case Candidate::Mul: 610 // C = Basis + Bump 611 if (BinaryOperator::isNeg(Bump)) { 612 // If Bump is a neg instruction, emit C = Basis - (-Bump). 613 Reduced = 614 Builder.CreateSub(Basis.Ins, BinaryOperator::getNegArgument(Bump)); 615 // We only use the negative argument of Bump, and Bump itself may be 616 // trivially dead. 617 RecursivelyDeleteTriviallyDeadInstructions(Bump); 618 } else { 619 // It's tempting to preserve nsw on Bump and/or Reduced. However, it's 620 // usually unsound, e.g., 621 // 622 // X = (-2 +nsw 1) *nsw INT_MAX 623 // Y = (-2 +nsw 3) *nsw INT_MAX 624 // => 625 // Y = X + 2 * INT_MAX 626 // 627 // Neither + and * in the resultant expression are nsw. 628 Reduced = Builder.CreateAdd(Basis.Ins, Bump); 629 } 630 break; 631 case Candidate::GEP: 632 { 633 Type *IntPtrTy = DL->getIntPtrType(C.Ins->getType()); 634 bool InBounds = cast<GetElementPtrInst>(C.Ins)->isInBounds(); 635 if (BumpWithUglyGEP) { 636 // C = (char *)Basis + Bump 637 unsigned AS = Basis.Ins->getType()->getPointerAddressSpace(); 638 Type *CharTy = Type::getInt8PtrTy(Basis.Ins->getContext(), AS); 639 Reduced = Builder.CreateBitCast(Basis.Ins, CharTy); 640 if (InBounds) 641 Reduced = 642 Builder.CreateInBoundsGEP(Builder.getInt8Ty(), Reduced, Bump); 643 else 644 Reduced = Builder.CreateGEP(Builder.getInt8Ty(), Reduced, Bump); 645 Reduced = Builder.CreateBitCast(Reduced, C.Ins->getType()); 646 } else { 647 // C = gep Basis, Bump 648 // Canonicalize bump to pointer size. 649 Bump = Builder.CreateSExtOrTrunc(Bump, IntPtrTy); 650 if (InBounds) 651 Reduced = Builder.CreateInBoundsGEP(nullptr, Basis.Ins, Bump); 652 else 653 Reduced = Builder.CreateGEP(nullptr, Basis.Ins, Bump); 654 } 655 } 656 break; 657 default: 658 llvm_unreachable("C.CandidateKind is invalid"); 659 }; 660 Reduced->takeName(C.Ins); 661 C.Ins->replaceAllUsesWith(Reduced); 662 // Unlink C.Ins so that we can skip other candidates also corresponding to 663 // C.Ins. The actual deletion is postponed to the end of runOnFunction. 664 C.Ins->removeFromParent(); 665 UnlinkedInstructions.push_back(C.Ins); 666 } 667 668 bool StraightLineStrengthReduce::runOnFunction(Function &F) { 669 if (skipFunction(F)) 670 return false; 671 672 TTI = &getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F); 673 DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree(); 674 SE = &getAnalysis<ScalarEvolutionWrapperPass>().getSE(); 675 // Traverse the dominator tree in the depth-first order. This order makes sure 676 // all bases of a candidate are in Candidates when we process it. 677 for (auto node = GraphTraits<DominatorTree *>::nodes_begin(DT); 678 node != GraphTraits<DominatorTree *>::nodes_end(DT); ++node) { 679 for (auto &I : *node->getBlock()) 680 allocateCandidatesAndFindBasis(&I); 681 } 682 683 // Rewrite candidates in the reverse depth-first order. This order makes sure 684 // a candidate being rewritten is not a basis for any other candidate. 685 while (!Candidates.empty()) { 686 const Candidate &C = Candidates.back(); 687 if (C.Basis != nullptr) { 688 rewriteCandidateWithBasis(C, *C.Basis); 689 } 690 Candidates.pop_back(); 691 } 692 693 // Delete all unlink instructions. 694 for (auto *UnlinkedInst : UnlinkedInstructions) { 695 for (unsigned I = 0, E = UnlinkedInst->getNumOperands(); I != E; ++I) { 696 Value *Op = UnlinkedInst->getOperand(I); 697 UnlinkedInst->setOperand(I, nullptr); 698 RecursivelyDeleteTriviallyDeadInstructions(Op); 699 } 700 delete UnlinkedInst; 701 } 702 bool Ret = !UnlinkedInstructions.empty(); 703 UnlinkedInstructions.clear(); 704 return Ret; 705 } 706