1 //===- Reassociate.cpp - Reassociate binary expressions -------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // This pass reassociates commutative expressions in an order that is designed 10 // to promote better constant propagation, GCSE, LICM, PRE, etc. 11 // 12 // For example: 4 + (x + 5) -> x + (4 + 5) 13 // 14 // In the implementation of this algorithm, constants are assigned rank = 0, 15 // function arguments are rank = 1, and other values are assigned ranks 16 // corresponding to the reverse post order traversal of current function 17 // (starting at 2), which effectively gives values in deep loops higher rank 18 // than values not in loops. 19 // 20 //===----------------------------------------------------------------------===// 21 22 #include "llvm/Transforms/Scalar/Reassociate.h" 23 #include "llvm/ADT/APFloat.h" 24 #include "llvm/ADT/APInt.h" 25 #include "llvm/ADT/DenseMap.h" 26 #include "llvm/ADT/PostOrderIterator.h" 27 #include "llvm/ADT/SmallPtrSet.h" 28 #include "llvm/ADT/SmallSet.h" 29 #include "llvm/ADT/SmallVector.h" 30 #include "llvm/ADT/Statistic.h" 31 #include "llvm/Analysis/BasicAliasAnalysis.h" 32 #include "llvm/Analysis/ConstantFolding.h" 33 #include "llvm/Analysis/GlobalsModRef.h" 34 #include "llvm/Analysis/ValueTracking.h" 35 #include "llvm/IR/Argument.h" 36 #include "llvm/IR/BasicBlock.h" 37 #include "llvm/IR/CFG.h" 38 #include "llvm/IR/Constant.h" 39 #include "llvm/IR/Constants.h" 40 #include "llvm/IR/Function.h" 41 #include "llvm/IR/IRBuilder.h" 42 #include "llvm/IR/InstrTypes.h" 43 #include "llvm/IR/Instruction.h" 44 #include "llvm/IR/Instructions.h" 45 #include "llvm/IR/Operator.h" 46 #include "llvm/IR/PassManager.h" 47 #include "llvm/IR/PatternMatch.h" 48 #include "llvm/IR/Type.h" 49 #include "llvm/IR/User.h" 50 #include "llvm/IR/Value.h" 51 #include "llvm/IR/ValueHandle.h" 52 #include "llvm/InitializePasses.h" 53 #include "llvm/Pass.h" 54 #include "llvm/Support/Casting.h" 55 #include "llvm/Support/Debug.h" 56 #include "llvm/Support/raw_ostream.h" 57 #include "llvm/Transforms/Scalar.h" 58 #include "llvm/Transforms/Utils/Local.h" 59 #include <algorithm> 60 #include <cassert> 61 #include <utility> 62 63 using namespace llvm; 64 using namespace reassociate; 65 using namespace PatternMatch; 66 67 #define DEBUG_TYPE "reassociate" 68 69 STATISTIC(NumChanged, "Number of insts reassociated"); 70 STATISTIC(NumAnnihil, "Number of expr tree annihilated"); 71 STATISTIC(NumFactor , "Number of multiplies factored"); 72 73 #ifndef NDEBUG 74 /// Print out the expression identified in the Ops list. 75 static void PrintOps(Instruction *I, const SmallVectorImpl<ValueEntry> &Ops) { 76 Module *M = I->getModule(); 77 dbgs() << Instruction::getOpcodeName(I->getOpcode()) << " " 78 << *Ops[0].Op->getType() << '\t'; 79 for (unsigned i = 0, e = Ops.size(); i != e; ++i) { 80 dbgs() << "[ "; 81 Ops[i].Op->printAsOperand(dbgs(), false, M); 82 dbgs() << ", #" << Ops[i].Rank << "] "; 83 } 84 } 85 #endif 86 87 /// Utility class representing a non-constant Xor-operand. We classify 88 /// non-constant Xor-Operands into two categories: 89 /// C1) The operand is in the form "X & C", where C is a constant and C != ~0 90 /// C2) 91 /// C2.1) The operand is in the form of "X | C", where C is a non-zero 92 /// constant. 93 /// C2.2) Any operand E which doesn't fall into C1 and C2.1, we view this 94 /// operand as "E | 0" 95 class llvm::reassociate::XorOpnd { 96 public: 97 XorOpnd(Value *V); 98 99 bool isInvalid() const { return SymbolicPart == nullptr; } 100 bool isOrExpr() const { return isOr; } 101 Value *getValue() const { return OrigVal; } 102 Value *getSymbolicPart() const { return SymbolicPart; } 103 unsigned getSymbolicRank() const { return SymbolicRank; } 104 const APInt &getConstPart() const { return ConstPart; } 105 106 void Invalidate() { SymbolicPart = OrigVal = nullptr; } 107 void setSymbolicRank(unsigned R) { SymbolicRank = R; } 108 109 private: 110 Value *OrigVal; 111 Value *SymbolicPart; 112 APInt ConstPart; 113 unsigned SymbolicRank; 114 bool isOr; 115 }; 116 117 XorOpnd::XorOpnd(Value *V) { 118 assert(!isa<ConstantInt>(V) && "No ConstantInt"); 119 OrigVal = V; 120 Instruction *I = dyn_cast<Instruction>(V); 121 SymbolicRank = 0; 122 123 if (I && (I->getOpcode() == Instruction::Or || 124 I->getOpcode() == Instruction::And)) { 125 Value *V0 = I->getOperand(0); 126 Value *V1 = I->getOperand(1); 127 const APInt *C; 128 if (match(V0, m_APInt(C))) 129 std::swap(V0, V1); 130 131 if (match(V1, m_APInt(C))) { 132 ConstPart = *C; 133 SymbolicPart = V0; 134 isOr = (I->getOpcode() == Instruction::Or); 135 return; 136 } 137 } 138 139 // view the operand as "V | 0" 140 SymbolicPart = V; 141 ConstPart = APInt::getZero(V->getType()->getScalarSizeInBits()); 142 isOr = true; 143 } 144 145 /// Return true if I is an instruction with the FastMathFlags that are needed 146 /// for general reassociation set. This is not the same as testing 147 /// Instruction::isAssociative() because it includes operations like fsub. 148 /// (This routine is only intended to be called for floating-point operations.) 149 static bool hasFPAssociativeFlags(Instruction *I) { 150 assert(I && isa<FPMathOperator>(I) && "Should only check FP ops"); 151 return I->hasAllowReassoc() && I->hasNoSignedZeros(); 152 } 153 154 /// Return true if V is an instruction of the specified opcode and if it 155 /// only has one use. 156 static BinaryOperator *isReassociableOp(Value *V, unsigned Opcode) { 157 auto *BO = dyn_cast<BinaryOperator>(V); 158 if (BO && BO->hasOneUse() && BO->getOpcode() == Opcode) 159 if (!isa<FPMathOperator>(BO) || hasFPAssociativeFlags(BO)) 160 return BO; 161 return nullptr; 162 } 163 164 static BinaryOperator *isReassociableOp(Value *V, unsigned Opcode1, 165 unsigned Opcode2) { 166 auto *BO = dyn_cast<BinaryOperator>(V); 167 if (BO && BO->hasOneUse() && 168 (BO->getOpcode() == Opcode1 || BO->getOpcode() == Opcode2)) 169 if (!isa<FPMathOperator>(BO) || hasFPAssociativeFlags(BO)) 170 return BO; 171 return nullptr; 172 } 173 174 void ReassociatePass::BuildRankMap(Function &F, 175 ReversePostOrderTraversal<Function*> &RPOT) { 176 unsigned Rank = 2; 177 178 // Assign distinct ranks to function arguments. 179 for (auto &Arg : F.args()) { 180 ValueRankMap[&Arg] = ++Rank; 181 LLVM_DEBUG(dbgs() << "Calculated Rank[" << Arg.getName() << "] = " << Rank 182 << "\n"); 183 } 184 185 // Traverse basic blocks in ReversePostOrder. 186 for (BasicBlock *BB : RPOT) { 187 unsigned BBRank = RankMap[BB] = ++Rank << 16; 188 189 // Walk the basic block, adding precomputed ranks for any instructions that 190 // we cannot move. This ensures that the ranks for these instructions are 191 // all different in the block. 192 for (Instruction &I : *BB) 193 if (mayHaveNonDefUseDependency(I)) 194 ValueRankMap[&I] = ++BBRank; 195 } 196 } 197 198 unsigned ReassociatePass::getRank(Value *V) { 199 Instruction *I = dyn_cast<Instruction>(V); 200 if (!I) { 201 if (isa<Argument>(V)) return ValueRankMap[V]; // Function argument. 202 return 0; // Otherwise it's a global or constant, rank 0. 203 } 204 205 if (unsigned Rank = ValueRankMap[I]) 206 return Rank; // Rank already known? 207 208 // If this is an expression, return the 1+MAX(rank(LHS), rank(RHS)) so that 209 // we can reassociate expressions for code motion! Since we do not recurse 210 // for PHI nodes, we cannot have infinite recursion here, because there 211 // cannot be loops in the value graph that do not go through PHI nodes. 212 unsigned Rank = 0, MaxRank = RankMap[I->getParent()]; 213 for (unsigned i = 0, e = I->getNumOperands(); i != e && Rank != MaxRank; ++i) 214 Rank = std::max(Rank, getRank(I->getOperand(i))); 215 216 // If this is a 'not' or 'neg' instruction, do not count it for rank. This 217 // assures us that X and ~X will have the same rank. 218 if (!match(I, m_Not(m_Value())) && !match(I, m_Neg(m_Value())) && 219 !match(I, m_FNeg(m_Value()))) 220 ++Rank; 221 222 LLVM_DEBUG(dbgs() << "Calculated Rank[" << V->getName() << "] = " << Rank 223 << "\n"); 224 225 return ValueRankMap[I] = Rank; 226 } 227 228 // Canonicalize constants to RHS. Otherwise, sort the operands by rank. 229 void ReassociatePass::canonicalizeOperands(Instruction *I) { 230 assert(isa<BinaryOperator>(I) && "Expected binary operator."); 231 assert(I->isCommutative() && "Expected commutative operator."); 232 233 Value *LHS = I->getOperand(0); 234 Value *RHS = I->getOperand(1); 235 if (LHS == RHS || isa<Constant>(RHS)) 236 return; 237 if (isa<Constant>(LHS) || getRank(RHS) < getRank(LHS)) 238 cast<BinaryOperator>(I)->swapOperands(); 239 } 240 241 static BinaryOperator *CreateAdd(Value *S1, Value *S2, const Twine &Name, 242 Instruction *InsertBefore, Value *FlagsOp) { 243 if (S1->getType()->isIntOrIntVectorTy()) 244 return BinaryOperator::CreateAdd(S1, S2, Name, InsertBefore); 245 else { 246 BinaryOperator *Res = 247 BinaryOperator::CreateFAdd(S1, S2, Name, InsertBefore); 248 Res->setFastMathFlags(cast<FPMathOperator>(FlagsOp)->getFastMathFlags()); 249 return Res; 250 } 251 } 252 253 static BinaryOperator *CreateMul(Value *S1, Value *S2, const Twine &Name, 254 Instruction *InsertBefore, Value *FlagsOp) { 255 if (S1->getType()->isIntOrIntVectorTy()) 256 return BinaryOperator::CreateMul(S1, S2, Name, InsertBefore); 257 else { 258 BinaryOperator *Res = 259 BinaryOperator::CreateFMul(S1, S2, Name, InsertBefore); 260 Res->setFastMathFlags(cast<FPMathOperator>(FlagsOp)->getFastMathFlags()); 261 return Res; 262 } 263 } 264 265 static Instruction *CreateNeg(Value *S1, const Twine &Name, 266 Instruction *InsertBefore, Value *FlagsOp) { 267 if (S1->getType()->isIntOrIntVectorTy()) 268 return BinaryOperator::CreateNeg(S1, Name, InsertBefore); 269 270 if (auto *FMFSource = dyn_cast<Instruction>(FlagsOp)) 271 return UnaryOperator::CreateFNegFMF(S1, FMFSource, Name, InsertBefore); 272 273 return UnaryOperator::CreateFNeg(S1, Name, InsertBefore); 274 } 275 276 /// Replace 0-X with X*-1. 277 static BinaryOperator *LowerNegateToMultiply(Instruction *Neg) { 278 assert((isa<UnaryOperator>(Neg) || isa<BinaryOperator>(Neg)) && 279 "Expected a Negate!"); 280 // FIXME: It's not safe to lower a unary FNeg into a FMul by -1.0. 281 unsigned OpNo = isa<BinaryOperator>(Neg) ? 1 : 0; 282 Type *Ty = Neg->getType(); 283 Constant *NegOne = Ty->isIntOrIntVectorTy() ? 284 ConstantInt::getAllOnesValue(Ty) : ConstantFP::get(Ty, -1.0); 285 286 BinaryOperator *Res = CreateMul(Neg->getOperand(OpNo), NegOne, "", Neg, Neg); 287 Neg->setOperand(OpNo, Constant::getNullValue(Ty)); // Drop use of op. 288 Res->takeName(Neg); 289 Neg->replaceAllUsesWith(Res); 290 Res->setDebugLoc(Neg->getDebugLoc()); 291 return Res; 292 } 293 294 /// Returns k such that lambda(2^Bitwidth) = 2^k, where lambda is the Carmichael 295 /// function. This means that x^(2^k) === 1 mod 2^Bitwidth for 296 /// every odd x, i.e. x^(2^k) = 1 for every odd x in Bitwidth-bit arithmetic. 297 /// Note that 0 <= k < Bitwidth, and if Bitwidth > 3 then x^(2^k) = 0 for every 298 /// even x in Bitwidth-bit arithmetic. 299 static unsigned CarmichaelShift(unsigned Bitwidth) { 300 if (Bitwidth < 3) 301 return Bitwidth - 1; 302 return Bitwidth - 2; 303 } 304 305 /// Add the extra weight 'RHS' to the existing weight 'LHS', 306 /// reducing the combined weight using any special properties of the operation. 307 /// The existing weight LHS represents the computation X op X op ... op X where 308 /// X occurs LHS times. The combined weight represents X op X op ... op X with 309 /// X occurring LHS + RHS times. If op is "Xor" for example then the combined 310 /// operation is equivalent to X if LHS + RHS is odd, or 0 if LHS + RHS is even; 311 /// the routine returns 1 in LHS in the first case, and 0 in LHS in the second. 312 static void IncorporateWeight(APInt &LHS, const APInt &RHS, unsigned Opcode) { 313 // If we were working with infinite precision arithmetic then the combined 314 // weight would be LHS + RHS. But we are using finite precision arithmetic, 315 // and the APInt sum LHS + RHS may not be correct if it wraps (it is correct 316 // for nilpotent operations and addition, but not for idempotent operations 317 // and multiplication), so it is important to correctly reduce the combined 318 // weight back into range if wrapping would be wrong. 319 320 // If RHS is zero then the weight didn't change. 321 if (RHS.isMinValue()) 322 return; 323 // If LHS is zero then the combined weight is RHS. 324 if (LHS.isMinValue()) { 325 LHS = RHS; 326 return; 327 } 328 // From this point on we know that neither LHS nor RHS is zero. 329 330 if (Instruction::isIdempotent(Opcode)) { 331 // Idempotent means X op X === X, so any non-zero weight is equivalent to a 332 // weight of 1. Keeping weights at zero or one also means that wrapping is 333 // not a problem. 334 assert(LHS == 1 && RHS == 1 && "Weights not reduced!"); 335 return; // Return a weight of 1. 336 } 337 if (Instruction::isNilpotent(Opcode)) { 338 // Nilpotent means X op X === 0, so reduce weights modulo 2. 339 assert(LHS == 1 && RHS == 1 && "Weights not reduced!"); 340 LHS = 0; // 1 + 1 === 0 modulo 2. 341 return; 342 } 343 if (Opcode == Instruction::Add || Opcode == Instruction::FAdd) { 344 // TODO: Reduce the weight by exploiting nsw/nuw? 345 LHS += RHS; 346 return; 347 } 348 349 assert((Opcode == Instruction::Mul || Opcode == Instruction::FMul) && 350 "Unknown associative operation!"); 351 unsigned Bitwidth = LHS.getBitWidth(); 352 // If CM is the Carmichael number then a weight W satisfying W >= CM+Bitwidth 353 // can be replaced with W-CM. That's because x^W=x^(W-CM) for every Bitwidth 354 // bit number x, since either x is odd in which case x^CM = 1, or x is even in 355 // which case both x^W and x^(W - CM) are zero. By subtracting off multiples 356 // of CM like this weights can always be reduced to the range [0, CM+Bitwidth) 357 // which by a happy accident means that they can always be represented using 358 // Bitwidth bits. 359 // TODO: Reduce the weight by exploiting nsw/nuw? (Could do much better than 360 // the Carmichael number). 361 if (Bitwidth > 3) { 362 /// CM - The value of Carmichael's lambda function. 363 APInt CM = APInt::getOneBitSet(Bitwidth, CarmichaelShift(Bitwidth)); 364 // Any weight W >= Threshold can be replaced with W - CM. 365 APInt Threshold = CM + Bitwidth; 366 assert(LHS.ult(Threshold) && RHS.ult(Threshold) && "Weights not reduced!"); 367 // For Bitwidth 4 or more the following sum does not overflow. 368 LHS += RHS; 369 while (LHS.uge(Threshold)) 370 LHS -= CM; 371 } else { 372 // To avoid problems with overflow do everything the same as above but using 373 // a larger type. 374 unsigned CM = 1U << CarmichaelShift(Bitwidth); 375 unsigned Threshold = CM + Bitwidth; 376 assert(LHS.getZExtValue() < Threshold && RHS.getZExtValue() < Threshold && 377 "Weights not reduced!"); 378 unsigned Total = LHS.getZExtValue() + RHS.getZExtValue(); 379 while (Total >= Threshold) 380 Total -= CM; 381 LHS = Total; 382 } 383 } 384 385 using RepeatedValue = std::pair<Value*, APInt>; 386 387 /// Given an associative binary expression, return the leaf 388 /// nodes in Ops along with their weights (how many times the leaf occurs). The 389 /// original expression is the same as 390 /// (Ops[0].first op Ops[0].first op ... Ops[0].first) <- Ops[0].second times 391 /// op 392 /// (Ops[1].first op Ops[1].first op ... Ops[1].first) <- Ops[1].second times 393 /// op 394 /// ... 395 /// op 396 /// (Ops[N].first op Ops[N].first op ... Ops[N].first) <- Ops[N].second times 397 /// 398 /// Note that the values Ops[0].first, ..., Ops[N].first are all distinct. 399 /// 400 /// This routine may modify the function, in which case it returns 'true'. The 401 /// changes it makes may well be destructive, changing the value computed by 'I' 402 /// to something completely different. Thus if the routine returns 'true' then 403 /// you MUST either replace I with a new expression computed from the Ops array, 404 /// or use RewriteExprTree to put the values back in. 405 /// 406 /// A leaf node is either not a binary operation of the same kind as the root 407 /// node 'I' (i.e. is not a binary operator at all, or is, but with a different 408 /// opcode), or is the same kind of binary operator but has a use which either 409 /// does not belong to the expression, or does belong to the expression but is 410 /// a leaf node. Every leaf node has at least one use that is a non-leaf node 411 /// of the expression, while for non-leaf nodes (except for the root 'I') every 412 /// use is a non-leaf node of the expression. 413 /// 414 /// For example: 415 /// expression graph node names 416 /// 417 /// + | I 418 /// / \ | 419 /// + + | A, B 420 /// / \ / \ | 421 /// * + * | C, D, E 422 /// / \ / \ / \ | 423 /// + * | F, G 424 /// 425 /// The leaf nodes are C, E, F and G. The Ops array will contain (maybe not in 426 /// that order) (C, 1), (E, 1), (F, 2), (G, 2). 427 /// 428 /// The expression is maximal: if some instruction is a binary operator of the 429 /// same kind as 'I', and all of its uses are non-leaf nodes of the expression, 430 /// then the instruction also belongs to the expression, is not a leaf node of 431 /// it, and its operands also belong to the expression (but may be leaf nodes). 432 /// 433 /// NOTE: This routine will set operands of non-leaf non-root nodes to undef in 434 /// order to ensure that every non-root node in the expression has *exactly one* 435 /// use by a non-leaf node of the expression. This destruction means that the 436 /// caller MUST either replace 'I' with a new expression or use something like 437 /// RewriteExprTree to put the values back in if the routine indicates that it 438 /// made a change by returning 'true'. 439 /// 440 /// In the above example either the right operand of A or the left operand of B 441 /// will be replaced by undef. If it is B's operand then this gives: 442 /// 443 /// + | I 444 /// / \ | 445 /// + + | A, B - operand of B replaced with undef 446 /// / \ \ | 447 /// * + * | C, D, E 448 /// / \ / \ / \ | 449 /// + * | F, G 450 /// 451 /// Note that such undef operands can only be reached by passing through 'I'. 452 /// For example, if you visit operands recursively starting from a leaf node 453 /// then you will never see such an undef operand unless you get back to 'I', 454 /// which requires passing through a phi node. 455 /// 456 /// Note that this routine may also mutate binary operators of the wrong type 457 /// that have all uses inside the expression (i.e. only used by non-leaf nodes 458 /// of the expression) if it can turn them into binary operators of the right 459 /// type and thus make the expression bigger. 460 static bool LinearizeExprTree(Instruction *I, 461 SmallVectorImpl<RepeatedValue> &Ops, 462 ReassociatePass::OrderedSet &ToRedo) { 463 assert((isa<UnaryOperator>(I) || isa<BinaryOperator>(I)) && 464 "Expected a UnaryOperator or BinaryOperator!"); 465 LLVM_DEBUG(dbgs() << "LINEARIZE: " << *I << '\n'); 466 unsigned Bitwidth = I->getType()->getScalarType()->getPrimitiveSizeInBits(); 467 unsigned Opcode = I->getOpcode(); 468 assert(I->isAssociative() && I->isCommutative() && 469 "Expected an associative and commutative operation!"); 470 471 // Visit all operands of the expression, keeping track of their weight (the 472 // number of paths from the expression root to the operand, or if you like 473 // the number of times that operand occurs in the linearized expression). 474 // For example, if I = X + A, where X = A + B, then I, X and B have weight 1 475 // while A has weight two. 476 477 // Worklist of non-leaf nodes (their operands are in the expression too) along 478 // with their weights, representing a certain number of paths to the operator. 479 // If an operator occurs in the worklist multiple times then we found multiple 480 // ways to get to it. 481 SmallVector<std::pair<Instruction*, APInt>, 8> Worklist; // (Op, Weight) 482 Worklist.push_back(std::make_pair(I, APInt(Bitwidth, 1))); 483 bool Changed = false; 484 485 // Leaves of the expression are values that either aren't the right kind of 486 // operation (eg: a constant, or a multiply in an add tree), or are, but have 487 // some uses that are not inside the expression. For example, in I = X + X, 488 // X = A + B, the value X has two uses (by I) that are in the expression. If 489 // X has any other uses, for example in a return instruction, then we consider 490 // X to be a leaf, and won't analyze it further. When we first visit a value, 491 // if it has more than one use then at first we conservatively consider it to 492 // be a leaf. Later, as the expression is explored, we may discover some more 493 // uses of the value from inside the expression. If all uses turn out to be 494 // from within the expression (and the value is a binary operator of the right 495 // kind) then the value is no longer considered to be a leaf, and its operands 496 // are explored. 497 498 // Leaves - Keeps track of the set of putative leaves as well as the number of 499 // paths to each leaf seen so far. 500 using LeafMap = DenseMap<Value *, APInt>; 501 LeafMap Leaves; // Leaf -> Total weight so far. 502 SmallVector<Value *, 8> LeafOrder; // Ensure deterministic leaf output order. 503 504 #ifndef NDEBUG 505 SmallPtrSet<Value *, 8> Visited; // For checking the iteration scheme. 506 #endif 507 while (!Worklist.empty()) { 508 std::pair<Instruction*, APInt> P = Worklist.pop_back_val(); 509 I = P.first; // We examine the operands of this binary operator. 510 511 for (unsigned OpIdx = 0; OpIdx < I->getNumOperands(); ++OpIdx) { // Visit operands. 512 Value *Op = I->getOperand(OpIdx); 513 APInt Weight = P.second; // Number of paths to this operand. 514 LLVM_DEBUG(dbgs() << "OPERAND: " << *Op << " (" << Weight << ")\n"); 515 assert(!Op->use_empty() && "No uses, so how did we get to it?!"); 516 517 // If this is a binary operation of the right kind with only one use then 518 // add its operands to the expression. 519 if (BinaryOperator *BO = isReassociableOp(Op, Opcode)) { 520 assert(Visited.insert(Op).second && "Not first visit!"); 521 LLVM_DEBUG(dbgs() << "DIRECT ADD: " << *Op << " (" << Weight << ")\n"); 522 Worklist.push_back(std::make_pair(BO, Weight)); 523 continue; 524 } 525 526 // Appears to be a leaf. Is the operand already in the set of leaves? 527 LeafMap::iterator It = Leaves.find(Op); 528 if (It == Leaves.end()) { 529 // Not in the leaf map. Must be the first time we saw this operand. 530 assert(Visited.insert(Op).second && "Not first visit!"); 531 if (!Op->hasOneUse()) { 532 // This value has uses not accounted for by the expression, so it is 533 // not safe to modify. Mark it as being a leaf. 534 LLVM_DEBUG(dbgs() 535 << "ADD USES LEAF: " << *Op << " (" << Weight << ")\n"); 536 LeafOrder.push_back(Op); 537 Leaves[Op] = Weight; 538 continue; 539 } 540 // No uses outside the expression, try morphing it. 541 } else { 542 // Already in the leaf map. 543 assert(It != Leaves.end() && Visited.count(Op) && 544 "In leaf map but not visited!"); 545 546 // Update the number of paths to the leaf. 547 IncorporateWeight(It->second, Weight, Opcode); 548 549 #if 0 // TODO: Re-enable once PR13021 is fixed. 550 // The leaf already has one use from inside the expression. As we want 551 // exactly one such use, drop this new use of the leaf. 552 assert(!Op->hasOneUse() && "Only one use, but we got here twice!"); 553 I->setOperand(OpIdx, UndefValue::get(I->getType())); 554 Changed = true; 555 556 // If the leaf is a binary operation of the right kind and we now see 557 // that its multiple original uses were in fact all by nodes belonging 558 // to the expression, then no longer consider it to be a leaf and add 559 // its operands to the expression. 560 if (BinaryOperator *BO = isReassociableOp(Op, Opcode)) { 561 LLVM_DEBUG(dbgs() << "UNLEAF: " << *Op << " (" << It->second << ")\n"); 562 Worklist.push_back(std::make_pair(BO, It->second)); 563 Leaves.erase(It); 564 continue; 565 } 566 #endif 567 568 // If we still have uses that are not accounted for by the expression 569 // then it is not safe to modify the value. 570 if (!Op->hasOneUse()) 571 continue; 572 573 // No uses outside the expression, try morphing it. 574 Weight = It->second; 575 Leaves.erase(It); // Since the value may be morphed below. 576 } 577 578 // At this point we have a value which, first of all, is not a binary 579 // expression of the right kind, and secondly, is only used inside the 580 // expression. This means that it can safely be modified. See if we 581 // can usefully morph it into an expression of the right kind. 582 assert((!isa<Instruction>(Op) || 583 cast<Instruction>(Op)->getOpcode() != Opcode 584 || (isa<FPMathOperator>(Op) && 585 !hasFPAssociativeFlags(cast<Instruction>(Op)))) && 586 "Should have been handled above!"); 587 assert(Op->hasOneUse() && "Has uses outside the expression tree!"); 588 589 // If this is a multiply expression, turn any internal negations into 590 // multiplies by -1 so they can be reassociated. Add any users of the 591 // newly created multiplication by -1 to the redo list, so any 592 // reassociation opportunities that are exposed will be reassociated 593 // further. 594 Instruction *Neg; 595 if (((Opcode == Instruction::Mul && match(Op, m_Neg(m_Value()))) || 596 (Opcode == Instruction::FMul && match(Op, m_FNeg(m_Value())))) && 597 match(Op, m_Instruction(Neg))) { 598 LLVM_DEBUG(dbgs() 599 << "MORPH LEAF: " << *Op << " (" << Weight << ") TO "); 600 Instruction *Mul = LowerNegateToMultiply(Neg); 601 LLVM_DEBUG(dbgs() << *Mul << '\n'); 602 Worklist.push_back(std::make_pair(Mul, Weight)); 603 for (User *U : Mul->users()) { 604 if (BinaryOperator *UserBO = dyn_cast<BinaryOperator>(U)) 605 ToRedo.insert(UserBO); 606 } 607 ToRedo.insert(Neg); 608 Changed = true; 609 continue; 610 } 611 612 // Failed to morph into an expression of the right type. This really is 613 // a leaf. 614 LLVM_DEBUG(dbgs() << "ADD LEAF: " << *Op << " (" << Weight << ")\n"); 615 assert(!isReassociableOp(Op, Opcode) && "Value was morphed?"); 616 LeafOrder.push_back(Op); 617 Leaves[Op] = Weight; 618 } 619 } 620 621 // The leaves, repeated according to their weights, represent the linearized 622 // form of the expression. 623 for (unsigned i = 0, e = LeafOrder.size(); i != e; ++i) { 624 Value *V = LeafOrder[i]; 625 LeafMap::iterator It = Leaves.find(V); 626 if (It == Leaves.end()) 627 // Node initially thought to be a leaf wasn't. 628 continue; 629 assert(!isReassociableOp(V, Opcode) && "Shouldn't be a leaf!"); 630 APInt Weight = It->second; 631 if (Weight.isMinValue()) 632 // Leaf already output or weight reduction eliminated it. 633 continue; 634 // Ensure the leaf is only output once. 635 It->second = 0; 636 Ops.push_back(std::make_pair(V, Weight)); 637 } 638 639 // For nilpotent operations or addition there may be no operands, for example 640 // because the expression was "X xor X" or consisted of 2^Bitwidth additions: 641 // in both cases the weight reduces to 0 causing the value to be skipped. 642 if (Ops.empty()) { 643 Constant *Identity = ConstantExpr::getBinOpIdentity(Opcode, I->getType()); 644 assert(Identity && "Associative operation without identity!"); 645 Ops.emplace_back(Identity, APInt(Bitwidth, 1)); 646 } 647 648 return Changed; 649 } 650 651 /// Now that the operands for this expression tree are 652 /// linearized and optimized, emit them in-order. 653 void ReassociatePass::RewriteExprTree(BinaryOperator *I, 654 SmallVectorImpl<ValueEntry> &Ops) { 655 assert(Ops.size() > 1 && "Single values should be used directly!"); 656 657 // Since our optimizations should never increase the number of operations, the 658 // new expression can usually be written reusing the existing binary operators 659 // from the original expression tree, without creating any new instructions, 660 // though the rewritten expression may have a completely different topology. 661 // We take care to not change anything if the new expression will be the same 662 // as the original. If more than trivial changes (like commuting operands) 663 // were made then we are obliged to clear out any optional subclass data like 664 // nsw flags. 665 666 /// NodesToRewrite - Nodes from the original expression available for writing 667 /// the new expression into. 668 SmallVector<BinaryOperator*, 8> NodesToRewrite; 669 unsigned Opcode = I->getOpcode(); 670 BinaryOperator *Op = I; 671 672 /// NotRewritable - The operands being written will be the leaves of the new 673 /// expression and must not be used as inner nodes (via NodesToRewrite) by 674 /// mistake. Inner nodes are always reassociable, and usually leaves are not 675 /// (if they were they would have been incorporated into the expression and so 676 /// would not be leaves), so most of the time there is no danger of this. But 677 /// in rare cases a leaf may become reassociable if an optimization kills uses 678 /// of it, or it may momentarily become reassociable during rewriting (below) 679 /// due it being removed as an operand of one of its uses. Ensure that misuse 680 /// of leaf nodes as inner nodes cannot occur by remembering all of the future 681 /// leaves and refusing to reuse any of them as inner nodes. 682 SmallPtrSet<Value*, 8> NotRewritable; 683 for (unsigned i = 0, e = Ops.size(); i != e; ++i) 684 NotRewritable.insert(Ops[i].Op); 685 686 // ExpressionChanged - Non-null if the rewritten expression differs from the 687 // original in some non-trivial way, requiring the clearing of optional flags. 688 // Flags are cleared from the operator in ExpressionChanged up to I inclusive. 689 BinaryOperator *ExpressionChanged = nullptr; 690 for (unsigned i = 0; ; ++i) { 691 // The last operation (which comes earliest in the IR) is special as both 692 // operands will come from Ops, rather than just one with the other being 693 // a subexpression. 694 if (i+2 == Ops.size()) { 695 Value *NewLHS = Ops[i].Op; 696 Value *NewRHS = Ops[i+1].Op; 697 Value *OldLHS = Op->getOperand(0); 698 Value *OldRHS = Op->getOperand(1); 699 700 if (NewLHS == OldLHS && NewRHS == OldRHS) 701 // Nothing changed, leave it alone. 702 break; 703 704 if (NewLHS == OldRHS && NewRHS == OldLHS) { 705 // The order of the operands was reversed. Swap them. 706 LLVM_DEBUG(dbgs() << "RA: " << *Op << '\n'); 707 Op->swapOperands(); 708 LLVM_DEBUG(dbgs() << "TO: " << *Op << '\n'); 709 MadeChange = true; 710 ++NumChanged; 711 break; 712 } 713 714 // The new operation differs non-trivially from the original. Overwrite 715 // the old operands with the new ones. 716 LLVM_DEBUG(dbgs() << "RA: " << *Op << '\n'); 717 if (NewLHS != OldLHS) { 718 BinaryOperator *BO = isReassociableOp(OldLHS, Opcode); 719 if (BO && !NotRewritable.count(BO)) 720 NodesToRewrite.push_back(BO); 721 Op->setOperand(0, NewLHS); 722 } 723 if (NewRHS != OldRHS) { 724 BinaryOperator *BO = isReassociableOp(OldRHS, Opcode); 725 if (BO && !NotRewritable.count(BO)) 726 NodesToRewrite.push_back(BO); 727 Op->setOperand(1, NewRHS); 728 } 729 LLVM_DEBUG(dbgs() << "TO: " << *Op << '\n'); 730 731 ExpressionChanged = Op; 732 MadeChange = true; 733 ++NumChanged; 734 735 break; 736 } 737 738 // Not the last operation. The left-hand side will be a sub-expression 739 // while the right-hand side will be the current element of Ops. 740 Value *NewRHS = Ops[i].Op; 741 if (NewRHS != Op->getOperand(1)) { 742 LLVM_DEBUG(dbgs() << "RA: " << *Op << '\n'); 743 if (NewRHS == Op->getOperand(0)) { 744 // The new right-hand side was already present as the left operand. If 745 // we are lucky then swapping the operands will sort out both of them. 746 Op->swapOperands(); 747 } else { 748 // Overwrite with the new right-hand side. 749 BinaryOperator *BO = isReassociableOp(Op->getOperand(1), Opcode); 750 if (BO && !NotRewritable.count(BO)) 751 NodesToRewrite.push_back(BO); 752 Op->setOperand(1, NewRHS); 753 ExpressionChanged = Op; 754 } 755 LLVM_DEBUG(dbgs() << "TO: " << *Op << '\n'); 756 MadeChange = true; 757 ++NumChanged; 758 } 759 760 // Now deal with the left-hand side. If this is already an operation node 761 // from the original expression then just rewrite the rest of the expression 762 // into it. 763 BinaryOperator *BO = isReassociableOp(Op->getOperand(0), Opcode); 764 if (BO && !NotRewritable.count(BO)) { 765 Op = BO; 766 continue; 767 } 768 769 // Otherwise, grab a spare node from the original expression and use that as 770 // the left-hand side. If there are no nodes left then the optimizers made 771 // an expression with more nodes than the original! This usually means that 772 // they did something stupid but it might mean that the problem was just too 773 // hard (finding the mimimal number of multiplications needed to realize a 774 // multiplication expression is NP-complete). Whatever the reason, smart or 775 // stupid, create a new node if there are none left. 776 BinaryOperator *NewOp; 777 if (NodesToRewrite.empty()) { 778 Constant *Undef = UndefValue::get(I->getType()); 779 NewOp = BinaryOperator::Create(Instruction::BinaryOps(Opcode), 780 Undef, Undef, "", I); 781 if (isa<FPMathOperator>(NewOp)) 782 NewOp->setFastMathFlags(I->getFastMathFlags()); 783 } else { 784 NewOp = NodesToRewrite.pop_back_val(); 785 } 786 787 LLVM_DEBUG(dbgs() << "RA: " << *Op << '\n'); 788 Op->setOperand(0, NewOp); 789 LLVM_DEBUG(dbgs() << "TO: " << *Op << '\n'); 790 ExpressionChanged = Op; 791 MadeChange = true; 792 ++NumChanged; 793 Op = NewOp; 794 } 795 796 // If the expression changed non-trivially then clear out all subclass data 797 // starting from the operator specified in ExpressionChanged, and compactify 798 // the operators to just before the expression root to guarantee that the 799 // expression tree is dominated by all of Ops. 800 if (ExpressionChanged) 801 do { 802 // Preserve FastMathFlags. 803 if (isa<FPMathOperator>(I)) { 804 FastMathFlags Flags = I->getFastMathFlags(); 805 ExpressionChanged->clearSubclassOptionalData(); 806 ExpressionChanged->setFastMathFlags(Flags); 807 } else 808 ExpressionChanged->clearSubclassOptionalData(); 809 810 if (ExpressionChanged == I) 811 break; 812 813 // Discard any debug info related to the expressions that has changed (we 814 // can leave debug infor related to the root, since the result of the 815 // expression tree should be the same even after reassociation). 816 replaceDbgUsesWithUndef(ExpressionChanged); 817 818 ExpressionChanged->moveBefore(I); 819 ExpressionChanged = cast<BinaryOperator>(*ExpressionChanged->user_begin()); 820 } while (true); 821 822 // Throw away any left over nodes from the original expression. 823 for (unsigned i = 0, e = NodesToRewrite.size(); i != e; ++i) 824 RedoInsts.insert(NodesToRewrite[i]); 825 } 826 827 /// Insert instructions before the instruction pointed to by BI, 828 /// that computes the negative version of the value specified. The negative 829 /// version of the value is returned, and BI is left pointing at the instruction 830 /// that should be processed next by the reassociation pass. 831 /// Also add intermediate instructions to the redo list that are modified while 832 /// pushing the negates through adds. These will be revisited to see if 833 /// additional opportunities have been exposed. 834 static Value *NegateValue(Value *V, Instruction *BI, 835 ReassociatePass::OrderedSet &ToRedo) { 836 if (auto *C = dyn_cast<Constant>(V)) { 837 const DataLayout &DL = BI->getModule()->getDataLayout(); 838 Constant *Res = C->getType()->isFPOrFPVectorTy() 839 ? ConstantFoldUnaryOpOperand(Instruction::FNeg, C, DL) 840 : ConstantExpr::getNeg(C); 841 if (Res) 842 return Res; 843 } 844 845 // We are trying to expose opportunity for reassociation. One of the things 846 // that we want to do to achieve this is to push a negation as deep into an 847 // expression chain as possible, to expose the add instructions. In practice, 848 // this means that we turn this: 849 // X = -(A+12+C+D) into X = -A + -12 + -C + -D = -12 + -A + -C + -D 850 // so that later, a: Y = 12+X could get reassociated with the -12 to eliminate 851 // the constants. We assume that instcombine will clean up the mess later if 852 // we introduce tons of unnecessary negation instructions. 853 // 854 if (BinaryOperator *I = 855 isReassociableOp(V, Instruction::Add, Instruction::FAdd)) { 856 // Push the negates through the add. 857 I->setOperand(0, NegateValue(I->getOperand(0), BI, ToRedo)); 858 I->setOperand(1, NegateValue(I->getOperand(1), BI, ToRedo)); 859 if (I->getOpcode() == Instruction::Add) { 860 I->setHasNoUnsignedWrap(false); 861 I->setHasNoSignedWrap(false); 862 } 863 864 // We must move the add instruction here, because the neg instructions do 865 // not dominate the old add instruction in general. By moving it, we are 866 // assured that the neg instructions we just inserted dominate the 867 // instruction we are about to insert after them. 868 // 869 I->moveBefore(BI); 870 I->setName(I->getName()+".neg"); 871 872 // Add the intermediate negates to the redo list as processing them later 873 // could expose more reassociating opportunities. 874 ToRedo.insert(I); 875 return I; 876 } 877 878 // Okay, we need to materialize a negated version of V with an instruction. 879 // Scan the use lists of V to see if we have one already. 880 for (User *U : V->users()) { 881 if (!match(U, m_Neg(m_Value())) && !match(U, m_FNeg(m_Value()))) 882 continue; 883 884 // We found one! Now we have to make sure that the definition dominates 885 // this use. We do this by moving it to the entry block (if it is a 886 // non-instruction value) or right after the definition. These negates will 887 // be zapped by reassociate later, so we don't need much finesse here. 888 Instruction *TheNeg = dyn_cast<Instruction>(U); 889 890 // We can't safely propagate a vector zero constant with poison/undef lanes. 891 Constant *C; 892 if (match(TheNeg, m_BinOp(m_Constant(C), m_Value())) && 893 C->containsUndefOrPoisonElement()) 894 continue; 895 896 // Verify that the negate is in this function, V might be a constant expr. 897 if (!TheNeg || 898 TheNeg->getParent()->getParent() != BI->getParent()->getParent()) 899 continue; 900 901 Instruction *InsertPt; 902 if (Instruction *InstInput = dyn_cast<Instruction>(V)) { 903 InsertPt = InstInput->getInsertionPointAfterDef(); 904 if (!InsertPt) 905 continue; 906 } else { 907 InsertPt = &*TheNeg->getFunction()->getEntryBlock().begin(); 908 } 909 910 TheNeg->moveBefore(InsertPt); 911 if (TheNeg->getOpcode() == Instruction::Sub) { 912 TheNeg->setHasNoUnsignedWrap(false); 913 TheNeg->setHasNoSignedWrap(false); 914 } else { 915 TheNeg->andIRFlags(BI); 916 } 917 ToRedo.insert(TheNeg); 918 return TheNeg; 919 } 920 921 // Insert a 'neg' instruction that subtracts the value from zero to get the 922 // negation. 923 Instruction *NewNeg = CreateNeg(V, V->getName() + ".neg", BI, BI); 924 ToRedo.insert(NewNeg); 925 return NewNeg; 926 } 927 928 // See if this `or` looks like an load widening reduction, i.e. that it 929 // consists of an `or`/`shl`/`zext`/`load` nodes only. Note that we don't 930 // ensure that the pattern is *really* a load widening reduction, 931 // we do not ensure that it can really be replaced with a widened load, 932 // only that it mostly looks like one. 933 static bool isLoadCombineCandidate(Instruction *Or) { 934 SmallVector<Instruction *, 8> Worklist; 935 SmallSet<Instruction *, 8> Visited; 936 937 auto Enqueue = [&](Value *V) { 938 auto *I = dyn_cast<Instruction>(V); 939 // Each node of an `or` reduction must be an instruction, 940 if (!I) 941 return false; // Node is certainly not part of an `or` load reduction. 942 // Only process instructions we have never processed before. 943 if (Visited.insert(I).second) 944 Worklist.emplace_back(I); 945 return true; // Will need to look at parent nodes. 946 }; 947 948 if (!Enqueue(Or)) 949 return false; // Not an `or` reduction pattern. 950 951 while (!Worklist.empty()) { 952 auto *I = Worklist.pop_back_val(); 953 954 // Okay, which instruction is this node? 955 switch (I->getOpcode()) { 956 case Instruction::Or: 957 // Got an `or` node. That's fine, just recurse into it's operands. 958 for (Value *Op : I->operands()) 959 if (!Enqueue(Op)) 960 return false; // Not an `or` reduction pattern. 961 continue; 962 963 case Instruction::Shl: 964 case Instruction::ZExt: 965 // `shl`/`zext` nodes are fine, just recurse into their base operand. 966 if (!Enqueue(I->getOperand(0))) 967 return false; // Not an `or` reduction pattern. 968 continue; 969 970 case Instruction::Load: 971 // Perfect, `load` node means we've reached an edge of the graph. 972 continue; 973 974 default: // Unknown node. 975 return false; // Not an `or` reduction pattern. 976 } 977 } 978 979 return true; 980 } 981 982 /// Return true if it may be profitable to convert this (X|Y) into (X+Y). 983 static bool shouldConvertOrWithNoCommonBitsToAdd(Instruction *Or) { 984 // Don't bother to convert this up unless either the LHS is an associable add 985 // or subtract or mul or if this is only used by one of the above. 986 // This is only a compile-time improvement, it is not needed for correctness! 987 auto isInteresting = [](Value *V) { 988 for (auto Op : {Instruction::Add, Instruction::Sub, Instruction::Mul, 989 Instruction::Shl}) 990 if (isReassociableOp(V, Op)) 991 return true; 992 return false; 993 }; 994 995 if (any_of(Or->operands(), isInteresting)) 996 return true; 997 998 Value *VB = Or->user_back(); 999 if (Or->hasOneUse() && isInteresting(VB)) 1000 return true; 1001 1002 return false; 1003 } 1004 1005 /// If we have (X|Y), and iff X and Y have no common bits set, 1006 /// transform this into (X+Y) to allow arithmetics reassociation. 1007 static BinaryOperator *convertOrWithNoCommonBitsToAdd(Instruction *Or) { 1008 // Convert an or into an add. 1009 BinaryOperator *New = 1010 CreateAdd(Or->getOperand(0), Or->getOperand(1), "", Or, Or); 1011 New->setHasNoSignedWrap(); 1012 New->setHasNoUnsignedWrap(); 1013 New->takeName(Or); 1014 1015 // Everyone now refers to the add instruction. 1016 Or->replaceAllUsesWith(New); 1017 New->setDebugLoc(Or->getDebugLoc()); 1018 1019 LLVM_DEBUG(dbgs() << "Converted or into an add: " << *New << '\n'); 1020 return New; 1021 } 1022 1023 /// Return true if we should break up this subtract of X-Y into (X + -Y). 1024 static bool ShouldBreakUpSubtract(Instruction *Sub) { 1025 // If this is a negation, we can't split it up! 1026 if (match(Sub, m_Neg(m_Value())) || match(Sub, m_FNeg(m_Value()))) 1027 return false; 1028 1029 // Don't breakup X - undef. 1030 if (isa<UndefValue>(Sub->getOperand(1))) 1031 return false; 1032 1033 // Don't bother to break this up unless either the LHS is an associable add or 1034 // subtract or if this is only used by one. 1035 Value *V0 = Sub->getOperand(0); 1036 if (isReassociableOp(V0, Instruction::Add, Instruction::FAdd) || 1037 isReassociableOp(V0, Instruction::Sub, Instruction::FSub)) 1038 return true; 1039 Value *V1 = Sub->getOperand(1); 1040 if (isReassociableOp(V1, Instruction::Add, Instruction::FAdd) || 1041 isReassociableOp(V1, Instruction::Sub, Instruction::FSub)) 1042 return true; 1043 Value *VB = Sub->user_back(); 1044 if (Sub->hasOneUse() && 1045 (isReassociableOp(VB, Instruction::Add, Instruction::FAdd) || 1046 isReassociableOp(VB, Instruction::Sub, Instruction::FSub))) 1047 return true; 1048 1049 return false; 1050 } 1051 1052 /// If we have (X-Y), and if either X is an add, or if this is only used by an 1053 /// add, transform this into (X+(0-Y)) to promote better reassociation. 1054 static BinaryOperator *BreakUpSubtract(Instruction *Sub, 1055 ReassociatePass::OrderedSet &ToRedo) { 1056 // Convert a subtract into an add and a neg instruction. This allows sub 1057 // instructions to be commuted with other add instructions. 1058 // 1059 // Calculate the negative value of Operand 1 of the sub instruction, 1060 // and set it as the RHS of the add instruction we just made. 1061 Value *NegVal = NegateValue(Sub->getOperand(1), Sub, ToRedo); 1062 BinaryOperator *New = CreateAdd(Sub->getOperand(0), NegVal, "", Sub, Sub); 1063 Sub->setOperand(0, Constant::getNullValue(Sub->getType())); // Drop use of op. 1064 Sub->setOperand(1, Constant::getNullValue(Sub->getType())); // Drop use of op. 1065 New->takeName(Sub); 1066 1067 // Everyone now refers to the add instruction. 1068 Sub->replaceAllUsesWith(New); 1069 New->setDebugLoc(Sub->getDebugLoc()); 1070 1071 LLVM_DEBUG(dbgs() << "Negated: " << *New << '\n'); 1072 return New; 1073 } 1074 1075 /// If this is a shift of a reassociable multiply or is used by one, change 1076 /// this into a multiply by a constant to assist with further reassociation. 1077 static BinaryOperator *ConvertShiftToMul(Instruction *Shl) { 1078 Constant *MulCst = ConstantInt::get(Shl->getType(), 1); 1079 auto *SA = cast<ConstantInt>(Shl->getOperand(1)); 1080 MulCst = ConstantExpr::getShl(MulCst, SA); 1081 1082 BinaryOperator *Mul = 1083 BinaryOperator::CreateMul(Shl->getOperand(0), MulCst, "", Shl); 1084 Shl->setOperand(0, PoisonValue::get(Shl->getType())); // Drop use of op. 1085 Mul->takeName(Shl); 1086 1087 // Everyone now refers to the mul instruction. 1088 Shl->replaceAllUsesWith(Mul); 1089 Mul->setDebugLoc(Shl->getDebugLoc()); 1090 1091 // We can safely preserve the nuw flag in all cases. It's also safe to turn a 1092 // nuw nsw shl into a nuw nsw mul. However, nsw in isolation requires special 1093 // handling. It can be preserved as long as we're not left shifting by 1094 // bitwidth - 1. 1095 bool NSW = cast<BinaryOperator>(Shl)->hasNoSignedWrap(); 1096 bool NUW = cast<BinaryOperator>(Shl)->hasNoUnsignedWrap(); 1097 unsigned BitWidth = Shl->getType()->getIntegerBitWidth(); 1098 if (NSW && (NUW || SA->getValue().ult(BitWidth - 1))) 1099 Mul->setHasNoSignedWrap(true); 1100 Mul->setHasNoUnsignedWrap(NUW); 1101 return Mul; 1102 } 1103 1104 /// Scan backwards and forwards among values with the same rank as element i 1105 /// to see if X exists. If X does not exist, return i. This is useful when 1106 /// scanning for 'x' when we see '-x' because they both get the same rank. 1107 static unsigned FindInOperandList(const SmallVectorImpl<ValueEntry> &Ops, 1108 unsigned i, Value *X) { 1109 unsigned XRank = Ops[i].Rank; 1110 unsigned e = Ops.size(); 1111 for (unsigned j = i+1; j != e && Ops[j].Rank == XRank; ++j) { 1112 if (Ops[j].Op == X) 1113 return j; 1114 if (Instruction *I1 = dyn_cast<Instruction>(Ops[j].Op)) 1115 if (Instruction *I2 = dyn_cast<Instruction>(X)) 1116 if (I1->isIdenticalTo(I2)) 1117 return j; 1118 } 1119 // Scan backwards. 1120 for (unsigned j = i-1; j != ~0U && Ops[j].Rank == XRank; --j) { 1121 if (Ops[j].Op == X) 1122 return j; 1123 if (Instruction *I1 = dyn_cast<Instruction>(Ops[j].Op)) 1124 if (Instruction *I2 = dyn_cast<Instruction>(X)) 1125 if (I1->isIdenticalTo(I2)) 1126 return j; 1127 } 1128 return i; 1129 } 1130 1131 /// Emit a tree of add instructions, summing Ops together 1132 /// and returning the result. Insert the tree before I. 1133 static Value *EmitAddTreeOfValues(Instruction *I, 1134 SmallVectorImpl<WeakTrackingVH> &Ops) { 1135 if (Ops.size() == 1) return Ops.back(); 1136 1137 Value *V1 = Ops.pop_back_val(); 1138 Value *V2 = EmitAddTreeOfValues(I, Ops); 1139 return CreateAdd(V2, V1, "reass.add", I, I); 1140 } 1141 1142 /// If V is an expression tree that is a multiplication sequence, 1143 /// and if this sequence contains a multiply by Factor, 1144 /// remove Factor from the tree and return the new tree. 1145 Value *ReassociatePass::RemoveFactorFromExpression(Value *V, Value *Factor) { 1146 BinaryOperator *BO = isReassociableOp(V, Instruction::Mul, Instruction::FMul); 1147 if (!BO) 1148 return nullptr; 1149 1150 SmallVector<RepeatedValue, 8> Tree; 1151 MadeChange |= LinearizeExprTree(BO, Tree, RedoInsts); 1152 SmallVector<ValueEntry, 8> Factors; 1153 Factors.reserve(Tree.size()); 1154 for (unsigned i = 0, e = Tree.size(); i != e; ++i) { 1155 RepeatedValue E = Tree[i]; 1156 Factors.append(E.second.getZExtValue(), 1157 ValueEntry(getRank(E.first), E.first)); 1158 } 1159 1160 bool FoundFactor = false; 1161 bool NeedsNegate = false; 1162 for (unsigned i = 0, e = Factors.size(); i != e; ++i) { 1163 if (Factors[i].Op == Factor) { 1164 FoundFactor = true; 1165 Factors.erase(Factors.begin()+i); 1166 break; 1167 } 1168 1169 // If this is a negative version of this factor, remove it. 1170 if (ConstantInt *FC1 = dyn_cast<ConstantInt>(Factor)) { 1171 if (ConstantInt *FC2 = dyn_cast<ConstantInt>(Factors[i].Op)) 1172 if (FC1->getValue() == -FC2->getValue()) { 1173 FoundFactor = NeedsNegate = true; 1174 Factors.erase(Factors.begin()+i); 1175 break; 1176 } 1177 } else if (ConstantFP *FC1 = dyn_cast<ConstantFP>(Factor)) { 1178 if (ConstantFP *FC2 = dyn_cast<ConstantFP>(Factors[i].Op)) { 1179 const APFloat &F1 = FC1->getValueAPF(); 1180 APFloat F2(FC2->getValueAPF()); 1181 F2.changeSign(); 1182 if (F1 == F2) { 1183 FoundFactor = NeedsNegate = true; 1184 Factors.erase(Factors.begin() + i); 1185 break; 1186 } 1187 } 1188 } 1189 } 1190 1191 if (!FoundFactor) { 1192 // Make sure to restore the operands to the expression tree. 1193 RewriteExprTree(BO, Factors); 1194 return nullptr; 1195 } 1196 1197 BasicBlock::iterator InsertPt = ++BO->getIterator(); 1198 1199 // If this was just a single multiply, remove the multiply and return the only 1200 // remaining operand. 1201 if (Factors.size() == 1) { 1202 RedoInsts.insert(BO); 1203 V = Factors[0].Op; 1204 } else { 1205 RewriteExprTree(BO, Factors); 1206 V = BO; 1207 } 1208 1209 if (NeedsNegate) 1210 V = CreateNeg(V, "neg", &*InsertPt, BO); 1211 1212 return V; 1213 } 1214 1215 /// If V is a single-use multiply, recursively add its operands as factors, 1216 /// otherwise add V to the list of factors. 1217 /// 1218 /// Ops is the top-level list of add operands we're trying to factor. 1219 static void FindSingleUseMultiplyFactors(Value *V, 1220 SmallVectorImpl<Value*> &Factors) { 1221 BinaryOperator *BO = isReassociableOp(V, Instruction::Mul, Instruction::FMul); 1222 if (!BO) { 1223 Factors.push_back(V); 1224 return; 1225 } 1226 1227 // Otherwise, add the LHS and RHS to the list of factors. 1228 FindSingleUseMultiplyFactors(BO->getOperand(1), Factors); 1229 FindSingleUseMultiplyFactors(BO->getOperand(0), Factors); 1230 } 1231 1232 /// Optimize a series of operands to an 'and', 'or', or 'xor' instruction. 1233 /// This optimizes based on identities. If it can be reduced to a single Value, 1234 /// it is returned, otherwise the Ops list is mutated as necessary. 1235 static Value *OptimizeAndOrXor(unsigned Opcode, 1236 SmallVectorImpl<ValueEntry> &Ops) { 1237 // Scan the operand lists looking for X and ~X pairs, along with X,X pairs. 1238 // If we find any, we can simplify the expression. X&~X == 0, X|~X == -1. 1239 for (unsigned i = 0, e = Ops.size(); i != e; ++i) { 1240 // First, check for X and ~X in the operand list. 1241 assert(i < Ops.size()); 1242 Value *X; 1243 if (match(Ops[i].Op, m_Not(m_Value(X)))) { // Cannot occur for ^. 1244 unsigned FoundX = FindInOperandList(Ops, i, X); 1245 if (FoundX != i) { 1246 if (Opcode == Instruction::And) // ...&X&~X = 0 1247 return Constant::getNullValue(X->getType()); 1248 1249 if (Opcode == Instruction::Or) // ...|X|~X = -1 1250 return Constant::getAllOnesValue(X->getType()); 1251 } 1252 } 1253 1254 // Next, check for duplicate pairs of values, which we assume are next to 1255 // each other, due to our sorting criteria. 1256 assert(i < Ops.size()); 1257 if (i+1 != Ops.size() && Ops[i+1].Op == Ops[i].Op) { 1258 if (Opcode == Instruction::And || Opcode == Instruction::Or) { 1259 // Drop duplicate values for And and Or. 1260 Ops.erase(Ops.begin()+i); 1261 --i; --e; 1262 ++NumAnnihil; 1263 continue; 1264 } 1265 1266 // Drop pairs of values for Xor. 1267 assert(Opcode == Instruction::Xor); 1268 if (e == 2) 1269 return Constant::getNullValue(Ops[0].Op->getType()); 1270 1271 // Y ^ X^X -> Y 1272 Ops.erase(Ops.begin()+i, Ops.begin()+i+2); 1273 i -= 1; e -= 2; 1274 ++NumAnnihil; 1275 } 1276 } 1277 return nullptr; 1278 } 1279 1280 /// Helper function of CombineXorOpnd(). It creates a bitwise-and 1281 /// instruction with the given two operands, and return the resulting 1282 /// instruction. There are two special cases: 1) if the constant operand is 0, 1283 /// it will return NULL. 2) if the constant is ~0, the symbolic operand will 1284 /// be returned. 1285 static Value *createAndInstr(Instruction *InsertBefore, Value *Opnd, 1286 const APInt &ConstOpnd) { 1287 if (ConstOpnd.isZero()) 1288 return nullptr; 1289 1290 if (ConstOpnd.isAllOnes()) 1291 return Opnd; 1292 1293 Instruction *I = BinaryOperator::CreateAnd( 1294 Opnd, ConstantInt::get(Opnd->getType(), ConstOpnd), "and.ra", 1295 InsertBefore); 1296 I->setDebugLoc(InsertBefore->getDebugLoc()); 1297 return I; 1298 } 1299 1300 // Helper function of OptimizeXor(). It tries to simplify "Opnd1 ^ ConstOpnd" 1301 // into "R ^ C", where C would be 0, and R is a symbolic value. 1302 // 1303 // If it was successful, true is returned, and the "R" and "C" is returned 1304 // via "Res" and "ConstOpnd", respectively; otherwise, false is returned, 1305 // and both "Res" and "ConstOpnd" remain unchanged. 1306 bool ReassociatePass::CombineXorOpnd(Instruction *I, XorOpnd *Opnd1, 1307 APInt &ConstOpnd, Value *&Res) { 1308 // Xor-Rule 1: (x | c1) ^ c2 = (x | c1) ^ (c1 ^ c1) ^ c2 1309 // = ((x | c1) ^ c1) ^ (c1 ^ c2) 1310 // = (x & ~c1) ^ (c1 ^ c2) 1311 // It is useful only when c1 == c2. 1312 if (!Opnd1->isOrExpr() || Opnd1->getConstPart().isZero()) 1313 return false; 1314 1315 if (!Opnd1->getValue()->hasOneUse()) 1316 return false; 1317 1318 const APInt &C1 = Opnd1->getConstPart(); 1319 if (C1 != ConstOpnd) 1320 return false; 1321 1322 Value *X = Opnd1->getSymbolicPart(); 1323 Res = createAndInstr(I, X, ~C1); 1324 // ConstOpnd was C2, now C1 ^ C2. 1325 ConstOpnd ^= C1; 1326 1327 if (Instruction *T = dyn_cast<Instruction>(Opnd1->getValue())) 1328 RedoInsts.insert(T); 1329 return true; 1330 } 1331 1332 // Helper function of OptimizeXor(). It tries to simplify 1333 // "Opnd1 ^ Opnd2 ^ ConstOpnd" into "R ^ C", where C would be 0, and R is a 1334 // symbolic value. 1335 // 1336 // If it was successful, true is returned, and the "R" and "C" is returned 1337 // via "Res" and "ConstOpnd", respectively (If the entire expression is 1338 // evaluated to a constant, the Res is set to NULL); otherwise, false is 1339 // returned, and both "Res" and "ConstOpnd" remain unchanged. 1340 bool ReassociatePass::CombineXorOpnd(Instruction *I, XorOpnd *Opnd1, 1341 XorOpnd *Opnd2, APInt &ConstOpnd, 1342 Value *&Res) { 1343 Value *X = Opnd1->getSymbolicPart(); 1344 if (X != Opnd2->getSymbolicPart()) 1345 return false; 1346 1347 // This many instruction become dead.(At least "Opnd1 ^ Opnd2" will die.) 1348 int DeadInstNum = 1; 1349 if (Opnd1->getValue()->hasOneUse()) 1350 DeadInstNum++; 1351 if (Opnd2->getValue()->hasOneUse()) 1352 DeadInstNum++; 1353 1354 // Xor-Rule 2: 1355 // (x | c1) ^ (x & c2) 1356 // = (x|c1) ^ (x&c2) ^ (c1 ^ c1) = ((x|c1) ^ c1) ^ (x & c2) ^ c1 1357 // = (x & ~c1) ^ (x & c2) ^ c1 // Xor-Rule 1 1358 // = (x & c3) ^ c1, where c3 = ~c1 ^ c2 // Xor-rule 3 1359 // 1360 if (Opnd1->isOrExpr() != Opnd2->isOrExpr()) { 1361 if (Opnd2->isOrExpr()) 1362 std::swap(Opnd1, Opnd2); 1363 1364 const APInt &C1 = Opnd1->getConstPart(); 1365 const APInt &C2 = Opnd2->getConstPart(); 1366 APInt C3((~C1) ^ C2); 1367 1368 // Do not increase code size! 1369 if (!C3.isZero() && !C3.isAllOnes()) { 1370 int NewInstNum = ConstOpnd.getBoolValue() ? 1 : 2; 1371 if (NewInstNum > DeadInstNum) 1372 return false; 1373 } 1374 1375 Res = createAndInstr(I, X, C3); 1376 ConstOpnd ^= C1; 1377 } else if (Opnd1->isOrExpr()) { 1378 // Xor-Rule 3: (x | c1) ^ (x | c2) = (x & c3) ^ c3 where c3 = c1 ^ c2 1379 // 1380 const APInt &C1 = Opnd1->getConstPart(); 1381 const APInt &C2 = Opnd2->getConstPart(); 1382 APInt C3 = C1 ^ C2; 1383 1384 // Do not increase code size 1385 if (!C3.isZero() && !C3.isAllOnes()) { 1386 int NewInstNum = ConstOpnd.getBoolValue() ? 1 : 2; 1387 if (NewInstNum > DeadInstNum) 1388 return false; 1389 } 1390 1391 Res = createAndInstr(I, X, C3); 1392 ConstOpnd ^= C3; 1393 } else { 1394 // Xor-Rule 4: (x & c1) ^ (x & c2) = (x & (c1^c2)) 1395 // 1396 const APInt &C1 = Opnd1->getConstPart(); 1397 const APInt &C2 = Opnd2->getConstPart(); 1398 APInt C3 = C1 ^ C2; 1399 Res = createAndInstr(I, X, C3); 1400 } 1401 1402 // Put the original operands in the Redo list; hope they will be deleted 1403 // as dead code. 1404 if (Instruction *T = dyn_cast<Instruction>(Opnd1->getValue())) 1405 RedoInsts.insert(T); 1406 if (Instruction *T = dyn_cast<Instruction>(Opnd2->getValue())) 1407 RedoInsts.insert(T); 1408 1409 return true; 1410 } 1411 1412 /// Optimize a series of operands to an 'xor' instruction. If it can be reduced 1413 /// to a single Value, it is returned, otherwise the Ops list is mutated as 1414 /// necessary. 1415 Value *ReassociatePass::OptimizeXor(Instruction *I, 1416 SmallVectorImpl<ValueEntry> &Ops) { 1417 if (Value *V = OptimizeAndOrXor(Instruction::Xor, Ops)) 1418 return V; 1419 1420 if (Ops.size() == 1) 1421 return nullptr; 1422 1423 SmallVector<XorOpnd, 8> Opnds; 1424 SmallVector<XorOpnd*, 8> OpndPtrs; 1425 Type *Ty = Ops[0].Op->getType(); 1426 APInt ConstOpnd(Ty->getScalarSizeInBits(), 0); 1427 1428 // Step 1: Convert ValueEntry to XorOpnd 1429 for (unsigned i = 0, e = Ops.size(); i != e; ++i) { 1430 Value *V = Ops[i].Op; 1431 const APInt *C; 1432 // TODO: Support non-splat vectors. 1433 if (match(V, m_APInt(C))) { 1434 ConstOpnd ^= *C; 1435 } else { 1436 XorOpnd O(V); 1437 O.setSymbolicRank(getRank(O.getSymbolicPart())); 1438 Opnds.push_back(O); 1439 } 1440 } 1441 1442 // NOTE: From this point on, do *NOT* add/delete element to/from "Opnds". 1443 // It would otherwise invalidate the "Opnds"'s iterator, and hence invalidate 1444 // the "OpndPtrs" as well. For the similar reason, do not fuse this loop 1445 // with the previous loop --- the iterator of the "Opnds" may be invalidated 1446 // when new elements are added to the vector. 1447 for (unsigned i = 0, e = Opnds.size(); i != e; ++i) 1448 OpndPtrs.push_back(&Opnds[i]); 1449 1450 // Step 2: Sort the Xor-Operands in a way such that the operands containing 1451 // the same symbolic value cluster together. For instance, the input operand 1452 // sequence ("x | 123", "y & 456", "x & 789") will be sorted into: 1453 // ("x | 123", "x & 789", "y & 456"). 1454 // 1455 // The purpose is twofold: 1456 // 1) Cluster together the operands sharing the same symbolic-value. 1457 // 2) Operand having smaller symbolic-value-rank is permuted earlier, which 1458 // could potentially shorten crital path, and expose more loop-invariants. 1459 // Note that values' rank are basically defined in RPO order (FIXME). 1460 // So, if Rank(X) < Rank(Y) < Rank(Z), it means X is defined earlier 1461 // than Y which is defined earlier than Z. Permute "x | 1", "Y & 2", 1462 // "z" in the order of X-Y-Z is better than any other orders. 1463 llvm::stable_sort(OpndPtrs, [](XorOpnd *LHS, XorOpnd *RHS) { 1464 return LHS->getSymbolicRank() < RHS->getSymbolicRank(); 1465 }); 1466 1467 // Step 3: Combine adjacent operands 1468 XorOpnd *PrevOpnd = nullptr; 1469 bool Changed = false; 1470 for (unsigned i = 0, e = Opnds.size(); i < e; i++) { 1471 XorOpnd *CurrOpnd = OpndPtrs[i]; 1472 // The combined value 1473 Value *CV; 1474 1475 // Step 3.1: Try simplifying "CurrOpnd ^ ConstOpnd" 1476 if (!ConstOpnd.isZero() && CombineXorOpnd(I, CurrOpnd, ConstOpnd, CV)) { 1477 Changed = true; 1478 if (CV) 1479 *CurrOpnd = XorOpnd(CV); 1480 else { 1481 CurrOpnd->Invalidate(); 1482 continue; 1483 } 1484 } 1485 1486 if (!PrevOpnd || CurrOpnd->getSymbolicPart() != PrevOpnd->getSymbolicPart()) { 1487 PrevOpnd = CurrOpnd; 1488 continue; 1489 } 1490 1491 // step 3.2: When previous and current operands share the same symbolic 1492 // value, try to simplify "PrevOpnd ^ CurrOpnd ^ ConstOpnd" 1493 if (CombineXorOpnd(I, CurrOpnd, PrevOpnd, ConstOpnd, CV)) { 1494 // Remove previous operand 1495 PrevOpnd->Invalidate(); 1496 if (CV) { 1497 *CurrOpnd = XorOpnd(CV); 1498 PrevOpnd = CurrOpnd; 1499 } else { 1500 CurrOpnd->Invalidate(); 1501 PrevOpnd = nullptr; 1502 } 1503 Changed = true; 1504 } 1505 } 1506 1507 // Step 4: Reassemble the Ops 1508 if (Changed) { 1509 Ops.clear(); 1510 for (unsigned int i = 0, e = Opnds.size(); i < e; i++) { 1511 XorOpnd &O = Opnds[i]; 1512 if (O.isInvalid()) 1513 continue; 1514 ValueEntry VE(getRank(O.getValue()), O.getValue()); 1515 Ops.push_back(VE); 1516 } 1517 if (!ConstOpnd.isZero()) { 1518 Value *C = ConstantInt::get(Ty, ConstOpnd); 1519 ValueEntry VE(getRank(C), C); 1520 Ops.push_back(VE); 1521 } 1522 unsigned Sz = Ops.size(); 1523 if (Sz == 1) 1524 return Ops.back().Op; 1525 if (Sz == 0) { 1526 assert(ConstOpnd.isZero()); 1527 return ConstantInt::get(Ty, ConstOpnd); 1528 } 1529 } 1530 1531 return nullptr; 1532 } 1533 1534 /// Optimize a series of operands to an 'add' instruction. This 1535 /// optimizes based on identities. If it can be reduced to a single Value, it 1536 /// is returned, otherwise the Ops list is mutated as necessary. 1537 Value *ReassociatePass::OptimizeAdd(Instruction *I, 1538 SmallVectorImpl<ValueEntry> &Ops) { 1539 // Scan the operand lists looking for X and -X pairs. If we find any, we 1540 // can simplify expressions like X+-X == 0 and X+~X ==-1. While we're at it, 1541 // scan for any 1542 // duplicates. We want to canonicalize Y+Y+Y+Z -> 3*Y+Z. 1543 1544 for (unsigned i = 0, e = Ops.size(); i != e; ++i) { 1545 Value *TheOp = Ops[i].Op; 1546 // Check to see if we've seen this operand before. If so, we factor all 1547 // instances of the operand together. Due to our sorting criteria, we know 1548 // that these need to be next to each other in the vector. 1549 if (i+1 != Ops.size() && Ops[i+1].Op == TheOp) { 1550 // Rescan the list, remove all instances of this operand from the expr. 1551 unsigned NumFound = 0; 1552 do { 1553 Ops.erase(Ops.begin()+i); 1554 ++NumFound; 1555 } while (i != Ops.size() && Ops[i].Op == TheOp); 1556 1557 LLVM_DEBUG(dbgs() << "\nFACTORING [" << NumFound << "]: " << *TheOp 1558 << '\n'); 1559 ++NumFactor; 1560 1561 // Insert a new multiply. 1562 Type *Ty = TheOp->getType(); 1563 Constant *C = Ty->isIntOrIntVectorTy() ? 1564 ConstantInt::get(Ty, NumFound) : ConstantFP::get(Ty, NumFound); 1565 Instruction *Mul = CreateMul(TheOp, C, "factor", I, I); 1566 1567 // Now that we have inserted a multiply, optimize it. This allows us to 1568 // handle cases that require multiple factoring steps, such as this: 1569 // (X*2) + (X*2) + (X*2) -> (X*2)*3 -> X*6 1570 RedoInsts.insert(Mul); 1571 1572 // If every add operand was a duplicate, return the multiply. 1573 if (Ops.empty()) 1574 return Mul; 1575 1576 // Otherwise, we had some input that didn't have the dupe, such as 1577 // "A + A + B" -> "A*2 + B". Add the new multiply to the list of 1578 // things being added by this operation. 1579 Ops.insert(Ops.begin(), ValueEntry(getRank(Mul), Mul)); 1580 1581 --i; 1582 e = Ops.size(); 1583 continue; 1584 } 1585 1586 // Check for X and -X or X and ~X in the operand list. 1587 Value *X; 1588 if (!match(TheOp, m_Neg(m_Value(X))) && !match(TheOp, m_Not(m_Value(X))) && 1589 !match(TheOp, m_FNeg(m_Value(X)))) 1590 continue; 1591 1592 unsigned FoundX = FindInOperandList(Ops, i, X); 1593 if (FoundX == i) 1594 continue; 1595 1596 // Remove X and -X from the operand list. 1597 if (Ops.size() == 2 && 1598 (match(TheOp, m_Neg(m_Value())) || match(TheOp, m_FNeg(m_Value())))) 1599 return Constant::getNullValue(X->getType()); 1600 1601 // Remove X and ~X from the operand list. 1602 if (Ops.size() == 2 && match(TheOp, m_Not(m_Value()))) 1603 return Constant::getAllOnesValue(X->getType()); 1604 1605 Ops.erase(Ops.begin()+i); 1606 if (i < FoundX) 1607 --FoundX; 1608 else 1609 --i; // Need to back up an extra one. 1610 Ops.erase(Ops.begin()+FoundX); 1611 ++NumAnnihil; 1612 --i; // Revisit element. 1613 e -= 2; // Removed two elements. 1614 1615 // if X and ~X we append -1 to the operand list. 1616 if (match(TheOp, m_Not(m_Value()))) { 1617 Value *V = Constant::getAllOnesValue(X->getType()); 1618 Ops.insert(Ops.end(), ValueEntry(getRank(V), V)); 1619 e += 1; 1620 } 1621 } 1622 1623 // Scan the operand list, checking to see if there are any common factors 1624 // between operands. Consider something like A*A+A*B*C+D. We would like to 1625 // reassociate this to A*(A+B*C)+D, which reduces the number of multiplies. 1626 // To efficiently find this, we count the number of times a factor occurs 1627 // for any ADD operands that are MULs. 1628 DenseMap<Value*, unsigned> FactorOccurrences; 1629 1630 // Keep track of each multiply we see, to avoid triggering on (X*4)+(X*4) 1631 // where they are actually the same multiply. 1632 unsigned MaxOcc = 0; 1633 Value *MaxOccVal = nullptr; 1634 for (unsigned i = 0, e = Ops.size(); i != e; ++i) { 1635 BinaryOperator *BOp = 1636 isReassociableOp(Ops[i].Op, Instruction::Mul, Instruction::FMul); 1637 if (!BOp) 1638 continue; 1639 1640 // Compute all of the factors of this added value. 1641 SmallVector<Value*, 8> Factors; 1642 FindSingleUseMultiplyFactors(BOp, Factors); 1643 assert(Factors.size() > 1 && "Bad linearize!"); 1644 1645 // Add one to FactorOccurrences for each unique factor in this op. 1646 SmallPtrSet<Value*, 8> Duplicates; 1647 for (unsigned i = 0, e = Factors.size(); i != e; ++i) { 1648 Value *Factor = Factors[i]; 1649 if (!Duplicates.insert(Factor).second) 1650 continue; 1651 1652 unsigned Occ = ++FactorOccurrences[Factor]; 1653 if (Occ > MaxOcc) { 1654 MaxOcc = Occ; 1655 MaxOccVal = Factor; 1656 } 1657 1658 // If Factor is a negative constant, add the negated value as a factor 1659 // because we can percolate the negate out. Watch for minint, which 1660 // cannot be positivified. 1661 if (ConstantInt *CI = dyn_cast<ConstantInt>(Factor)) { 1662 if (CI->isNegative() && !CI->isMinValue(true)) { 1663 Factor = ConstantInt::get(CI->getContext(), -CI->getValue()); 1664 if (!Duplicates.insert(Factor).second) 1665 continue; 1666 unsigned Occ = ++FactorOccurrences[Factor]; 1667 if (Occ > MaxOcc) { 1668 MaxOcc = Occ; 1669 MaxOccVal = Factor; 1670 } 1671 } 1672 } else if (ConstantFP *CF = dyn_cast<ConstantFP>(Factor)) { 1673 if (CF->isNegative()) { 1674 APFloat F(CF->getValueAPF()); 1675 F.changeSign(); 1676 Factor = ConstantFP::get(CF->getContext(), F); 1677 if (!Duplicates.insert(Factor).second) 1678 continue; 1679 unsigned Occ = ++FactorOccurrences[Factor]; 1680 if (Occ > MaxOcc) { 1681 MaxOcc = Occ; 1682 MaxOccVal = Factor; 1683 } 1684 } 1685 } 1686 } 1687 } 1688 1689 // If any factor occurred more than one time, we can pull it out. 1690 if (MaxOcc > 1) { 1691 LLVM_DEBUG(dbgs() << "\nFACTORING [" << MaxOcc << "]: " << *MaxOccVal 1692 << '\n'); 1693 ++NumFactor; 1694 1695 // Create a new instruction that uses the MaxOccVal twice. If we don't do 1696 // this, we could otherwise run into situations where removing a factor 1697 // from an expression will drop a use of maxocc, and this can cause 1698 // RemoveFactorFromExpression on successive values to behave differently. 1699 Instruction *DummyInst = 1700 I->getType()->isIntOrIntVectorTy() 1701 ? BinaryOperator::CreateAdd(MaxOccVal, MaxOccVal) 1702 : BinaryOperator::CreateFAdd(MaxOccVal, MaxOccVal); 1703 1704 SmallVector<WeakTrackingVH, 4> NewMulOps; 1705 for (unsigned i = 0; i != Ops.size(); ++i) { 1706 // Only try to remove factors from expressions we're allowed to. 1707 BinaryOperator *BOp = 1708 isReassociableOp(Ops[i].Op, Instruction::Mul, Instruction::FMul); 1709 if (!BOp) 1710 continue; 1711 1712 if (Value *V = RemoveFactorFromExpression(Ops[i].Op, MaxOccVal)) { 1713 // The factorized operand may occur several times. Convert them all in 1714 // one fell swoop. 1715 for (unsigned j = Ops.size(); j != i;) { 1716 --j; 1717 if (Ops[j].Op == Ops[i].Op) { 1718 NewMulOps.push_back(V); 1719 Ops.erase(Ops.begin()+j); 1720 } 1721 } 1722 --i; 1723 } 1724 } 1725 1726 // No need for extra uses anymore. 1727 DummyInst->deleteValue(); 1728 1729 unsigned NumAddedValues = NewMulOps.size(); 1730 Value *V = EmitAddTreeOfValues(I, NewMulOps); 1731 1732 // Now that we have inserted the add tree, optimize it. This allows us to 1733 // handle cases that require multiple factoring steps, such as this: 1734 // A*A*B + A*A*C --> A*(A*B+A*C) --> A*(A*(B+C)) 1735 assert(NumAddedValues > 1 && "Each occurrence should contribute a value"); 1736 (void)NumAddedValues; 1737 if (Instruction *VI = dyn_cast<Instruction>(V)) 1738 RedoInsts.insert(VI); 1739 1740 // Create the multiply. 1741 Instruction *V2 = CreateMul(V, MaxOccVal, "reass.mul", I, I); 1742 1743 // Rerun associate on the multiply in case the inner expression turned into 1744 // a multiply. We want to make sure that we keep things in canonical form. 1745 RedoInsts.insert(V2); 1746 1747 // If every add operand included the factor (e.g. "A*B + A*C"), then the 1748 // entire result expression is just the multiply "A*(B+C)". 1749 if (Ops.empty()) 1750 return V2; 1751 1752 // Otherwise, we had some input that didn't have the factor, such as 1753 // "A*B + A*C + D" -> "A*(B+C) + D". Add the new multiply to the list of 1754 // things being added by this operation. 1755 Ops.insert(Ops.begin(), ValueEntry(getRank(V2), V2)); 1756 } 1757 1758 return nullptr; 1759 } 1760 1761 /// Build up a vector of value/power pairs factoring a product. 1762 /// 1763 /// Given a series of multiplication operands, build a vector of factors and 1764 /// the powers each is raised to when forming the final product. Sort them in 1765 /// the order of descending power. 1766 /// 1767 /// (x*x) -> [(x, 2)] 1768 /// ((x*x)*x) -> [(x, 3)] 1769 /// ((((x*y)*x)*y)*x) -> [(x, 3), (y, 2)] 1770 /// 1771 /// \returns Whether any factors have a power greater than one. 1772 static bool collectMultiplyFactors(SmallVectorImpl<ValueEntry> &Ops, 1773 SmallVectorImpl<Factor> &Factors) { 1774 // FIXME: Have Ops be (ValueEntry, Multiplicity) pairs, simplifying this. 1775 // Compute the sum of powers of simplifiable factors. 1776 unsigned FactorPowerSum = 0; 1777 for (unsigned Idx = 1, Size = Ops.size(); Idx < Size; ++Idx) { 1778 Value *Op = Ops[Idx-1].Op; 1779 1780 // Count the number of occurrences of this value. 1781 unsigned Count = 1; 1782 for (; Idx < Size && Ops[Idx].Op == Op; ++Idx) 1783 ++Count; 1784 // Track for simplification all factors which occur 2 or more times. 1785 if (Count > 1) 1786 FactorPowerSum += Count; 1787 } 1788 1789 // We can only simplify factors if the sum of the powers of our simplifiable 1790 // factors is 4 or higher. When that is the case, we will *always* have 1791 // a simplification. This is an important invariant to prevent cyclicly 1792 // trying to simplify already minimal formations. 1793 if (FactorPowerSum < 4) 1794 return false; 1795 1796 // Now gather the simplifiable factors, removing them from Ops. 1797 FactorPowerSum = 0; 1798 for (unsigned Idx = 1; Idx < Ops.size(); ++Idx) { 1799 Value *Op = Ops[Idx-1].Op; 1800 1801 // Count the number of occurrences of this value. 1802 unsigned Count = 1; 1803 for (; Idx < Ops.size() && Ops[Idx].Op == Op; ++Idx) 1804 ++Count; 1805 if (Count == 1) 1806 continue; 1807 // Move an even number of occurrences to Factors. 1808 Count &= ~1U; 1809 Idx -= Count; 1810 FactorPowerSum += Count; 1811 Factors.push_back(Factor(Op, Count)); 1812 Ops.erase(Ops.begin()+Idx, Ops.begin()+Idx+Count); 1813 } 1814 1815 // None of the adjustments above should have reduced the sum of factor powers 1816 // below our mininum of '4'. 1817 assert(FactorPowerSum >= 4); 1818 1819 llvm::stable_sort(Factors, [](const Factor &LHS, const Factor &RHS) { 1820 return LHS.Power > RHS.Power; 1821 }); 1822 return true; 1823 } 1824 1825 /// Build a tree of multiplies, computing the product of Ops. 1826 static Value *buildMultiplyTree(IRBuilderBase &Builder, 1827 SmallVectorImpl<Value*> &Ops) { 1828 if (Ops.size() == 1) 1829 return Ops.back(); 1830 1831 Value *LHS = Ops.pop_back_val(); 1832 do { 1833 if (LHS->getType()->isIntOrIntVectorTy()) 1834 LHS = Builder.CreateMul(LHS, Ops.pop_back_val()); 1835 else 1836 LHS = Builder.CreateFMul(LHS, Ops.pop_back_val()); 1837 } while (!Ops.empty()); 1838 1839 return LHS; 1840 } 1841 1842 /// Build a minimal multiplication DAG for (a^x)*(b^y)*(c^z)*... 1843 /// 1844 /// Given a vector of values raised to various powers, where no two values are 1845 /// equal and the powers are sorted in decreasing order, compute the minimal 1846 /// DAG of multiplies to compute the final product, and return that product 1847 /// value. 1848 Value * 1849 ReassociatePass::buildMinimalMultiplyDAG(IRBuilderBase &Builder, 1850 SmallVectorImpl<Factor> &Factors) { 1851 assert(Factors[0].Power); 1852 SmallVector<Value *, 4> OuterProduct; 1853 for (unsigned LastIdx = 0, Idx = 1, Size = Factors.size(); 1854 Idx < Size && Factors[Idx].Power > 0; ++Idx) { 1855 if (Factors[Idx].Power != Factors[LastIdx].Power) { 1856 LastIdx = Idx; 1857 continue; 1858 } 1859 1860 // We want to multiply across all the factors with the same power so that 1861 // we can raise them to that power as a single entity. Build a mini tree 1862 // for that. 1863 SmallVector<Value *, 4> InnerProduct; 1864 InnerProduct.push_back(Factors[LastIdx].Base); 1865 do { 1866 InnerProduct.push_back(Factors[Idx].Base); 1867 ++Idx; 1868 } while (Idx < Size && Factors[Idx].Power == Factors[LastIdx].Power); 1869 1870 // Reset the base value of the first factor to the new expression tree. 1871 // We'll remove all the factors with the same power in a second pass. 1872 Value *M = Factors[LastIdx].Base = buildMultiplyTree(Builder, InnerProduct); 1873 if (Instruction *MI = dyn_cast<Instruction>(M)) 1874 RedoInsts.insert(MI); 1875 1876 LastIdx = Idx; 1877 } 1878 // Unique factors with equal powers -- we've folded them into the first one's 1879 // base. 1880 Factors.erase(std::unique(Factors.begin(), Factors.end(), 1881 [](const Factor &LHS, const Factor &RHS) { 1882 return LHS.Power == RHS.Power; 1883 }), 1884 Factors.end()); 1885 1886 // Iteratively collect the base of each factor with an add power into the 1887 // outer product, and halve each power in preparation for squaring the 1888 // expression. 1889 for (Factor &F : Factors) { 1890 if (F.Power & 1) 1891 OuterProduct.push_back(F.Base); 1892 F.Power >>= 1; 1893 } 1894 if (Factors[0].Power) { 1895 Value *SquareRoot = buildMinimalMultiplyDAG(Builder, Factors); 1896 OuterProduct.push_back(SquareRoot); 1897 OuterProduct.push_back(SquareRoot); 1898 } 1899 if (OuterProduct.size() == 1) 1900 return OuterProduct.front(); 1901 1902 Value *V = buildMultiplyTree(Builder, OuterProduct); 1903 return V; 1904 } 1905 1906 Value *ReassociatePass::OptimizeMul(BinaryOperator *I, 1907 SmallVectorImpl<ValueEntry> &Ops) { 1908 // We can only optimize the multiplies when there is a chain of more than 1909 // three, such that a balanced tree might require fewer total multiplies. 1910 if (Ops.size() < 4) 1911 return nullptr; 1912 1913 // Try to turn linear trees of multiplies without other uses of the 1914 // intermediate stages into minimal multiply DAGs with perfect sub-expression 1915 // re-use. 1916 SmallVector<Factor, 4> Factors; 1917 if (!collectMultiplyFactors(Ops, Factors)) 1918 return nullptr; // All distinct factors, so nothing left for us to do. 1919 1920 IRBuilder<> Builder(I); 1921 // The reassociate transformation for FP operations is performed only 1922 // if unsafe algebra is permitted by FastMathFlags. Propagate those flags 1923 // to the newly generated operations. 1924 if (auto FPI = dyn_cast<FPMathOperator>(I)) 1925 Builder.setFastMathFlags(FPI->getFastMathFlags()); 1926 1927 Value *V = buildMinimalMultiplyDAG(Builder, Factors); 1928 if (Ops.empty()) 1929 return V; 1930 1931 ValueEntry NewEntry = ValueEntry(getRank(V), V); 1932 Ops.insert(llvm::lower_bound(Ops, NewEntry), NewEntry); 1933 return nullptr; 1934 } 1935 1936 Value *ReassociatePass::OptimizeExpression(BinaryOperator *I, 1937 SmallVectorImpl<ValueEntry> &Ops) { 1938 // Now that we have the linearized expression tree, try to optimize it. 1939 // Start by folding any constants that we found. 1940 const DataLayout &DL = I->getModule()->getDataLayout(); 1941 Constant *Cst = nullptr; 1942 unsigned Opcode = I->getOpcode(); 1943 while (!Ops.empty()) { 1944 if (auto *C = dyn_cast<Constant>(Ops.back().Op)) { 1945 if (!Cst) { 1946 Ops.pop_back(); 1947 Cst = C; 1948 continue; 1949 } 1950 if (Constant *Res = ConstantFoldBinaryOpOperands(Opcode, C, Cst, DL)) { 1951 Ops.pop_back(); 1952 Cst = Res; 1953 continue; 1954 } 1955 } 1956 break; 1957 } 1958 // If there was nothing but constants then we are done. 1959 if (Ops.empty()) 1960 return Cst; 1961 1962 // Put the combined constant back at the end of the operand list, except if 1963 // there is no point. For example, an add of 0 gets dropped here, while a 1964 // multiplication by zero turns the whole expression into zero. 1965 if (Cst && Cst != ConstantExpr::getBinOpIdentity(Opcode, I->getType())) { 1966 if (Cst == ConstantExpr::getBinOpAbsorber(Opcode, I->getType())) 1967 return Cst; 1968 Ops.push_back(ValueEntry(0, Cst)); 1969 } 1970 1971 if (Ops.size() == 1) return Ops[0].Op; 1972 1973 // Handle destructive annihilation due to identities between elements in the 1974 // argument list here. 1975 unsigned NumOps = Ops.size(); 1976 switch (Opcode) { 1977 default: break; 1978 case Instruction::And: 1979 case Instruction::Or: 1980 if (Value *Result = OptimizeAndOrXor(Opcode, Ops)) 1981 return Result; 1982 break; 1983 1984 case Instruction::Xor: 1985 if (Value *Result = OptimizeXor(I, Ops)) 1986 return Result; 1987 break; 1988 1989 case Instruction::Add: 1990 case Instruction::FAdd: 1991 if (Value *Result = OptimizeAdd(I, Ops)) 1992 return Result; 1993 break; 1994 1995 case Instruction::Mul: 1996 case Instruction::FMul: 1997 if (Value *Result = OptimizeMul(I, Ops)) 1998 return Result; 1999 break; 2000 } 2001 2002 if (Ops.size() != NumOps) 2003 return OptimizeExpression(I, Ops); 2004 return nullptr; 2005 } 2006 2007 // Remove dead instructions and if any operands are trivially dead add them to 2008 // Insts so they will be removed as well. 2009 void ReassociatePass::RecursivelyEraseDeadInsts(Instruction *I, 2010 OrderedSet &Insts) { 2011 assert(isInstructionTriviallyDead(I) && "Trivially dead instructions only!"); 2012 SmallVector<Value *, 4> Ops(I->operands()); 2013 ValueRankMap.erase(I); 2014 Insts.remove(I); 2015 RedoInsts.remove(I); 2016 llvm::salvageDebugInfo(*I); 2017 I->eraseFromParent(); 2018 for (auto *Op : Ops) 2019 if (Instruction *OpInst = dyn_cast<Instruction>(Op)) 2020 if (OpInst->use_empty()) 2021 Insts.insert(OpInst); 2022 } 2023 2024 /// Zap the given instruction, adding interesting operands to the work list. 2025 void ReassociatePass::EraseInst(Instruction *I) { 2026 assert(isInstructionTriviallyDead(I) && "Trivially dead instructions only!"); 2027 LLVM_DEBUG(dbgs() << "Erasing dead inst: "; I->dump()); 2028 2029 SmallVector<Value *, 8> Ops(I->operands()); 2030 // Erase the dead instruction. 2031 ValueRankMap.erase(I); 2032 RedoInsts.remove(I); 2033 llvm::salvageDebugInfo(*I); 2034 I->eraseFromParent(); 2035 // Optimize its operands. 2036 SmallPtrSet<Instruction *, 8> Visited; // Detect self-referential nodes. 2037 for (unsigned i = 0, e = Ops.size(); i != e; ++i) 2038 if (Instruction *Op = dyn_cast<Instruction>(Ops[i])) { 2039 // If this is a node in an expression tree, climb to the expression root 2040 // and add that since that's where optimization actually happens. 2041 unsigned Opcode = Op->getOpcode(); 2042 while (Op->hasOneUse() && Op->user_back()->getOpcode() == Opcode && 2043 Visited.insert(Op).second) 2044 Op = Op->user_back(); 2045 2046 // The instruction we're going to push may be coming from a 2047 // dead block, and Reassociate skips the processing of unreachable 2048 // blocks because it's a waste of time and also because it can 2049 // lead to infinite loop due to LLVM's non-standard definition 2050 // of dominance. 2051 if (ValueRankMap.find(Op) != ValueRankMap.end()) 2052 RedoInsts.insert(Op); 2053 } 2054 2055 MadeChange = true; 2056 } 2057 2058 /// Recursively analyze an expression to build a list of instructions that have 2059 /// negative floating-point constant operands. The caller can then transform 2060 /// the list to create positive constants for better reassociation and CSE. 2061 static void getNegatibleInsts(Value *V, 2062 SmallVectorImpl<Instruction *> &Candidates) { 2063 // Handle only one-use instructions. Combining negations does not justify 2064 // replicating instructions. 2065 Instruction *I; 2066 if (!match(V, m_OneUse(m_Instruction(I)))) 2067 return; 2068 2069 // Handle expressions of multiplications and divisions. 2070 // TODO: This could look through floating-point casts. 2071 const APFloat *C; 2072 switch (I->getOpcode()) { 2073 case Instruction::FMul: 2074 // Not expecting non-canonical code here. Bail out and wait. 2075 if (match(I->getOperand(0), m_Constant())) 2076 break; 2077 2078 if (match(I->getOperand(1), m_APFloat(C)) && C->isNegative()) { 2079 Candidates.push_back(I); 2080 LLVM_DEBUG(dbgs() << "FMul with negative constant: " << *I << '\n'); 2081 } 2082 getNegatibleInsts(I->getOperand(0), Candidates); 2083 getNegatibleInsts(I->getOperand(1), Candidates); 2084 break; 2085 case Instruction::FDiv: 2086 // Not expecting non-canonical code here. Bail out and wait. 2087 if (match(I->getOperand(0), m_Constant()) && 2088 match(I->getOperand(1), m_Constant())) 2089 break; 2090 2091 if ((match(I->getOperand(0), m_APFloat(C)) && C->isNegative()) || 2092 (match(I->getOperand(1), m_APFloat(C)) && C->isNegative())) { 2093 Candidates.push_back(I); 2094 LLVM_DEBUG(dbgs() << "FDiv with negative constant: " << *I << '\n'); 2095 } 2096 getNegatibleInsts(I->getOperand(0), Candidates); 2097 getNegatibleInsts(I->getOperand(1), Candidates); 2098 break; 2099 default: 2100 break; 2101 } 2102 } 2103 2104 /// Given an fadd/fsub with an operand that is a one-use instruction 2105 /// (the fadd/fsub), try to change negative floating-point constants into 2106 /// positive constants to increase potential for reassociation and CSE. 2107 Instruction *ReassociatePass::canonicalizeNegFPConstantsForOp(Instruction *I, 2108 Instruction *Op, 2109 Value *OtherOp) { 2110 assert((I->getOpcode() == Instruction::FAdd || 2111 I->getOpcode() == Instruction::FSub) && "Expected fadd/fsub"); 2112 2113 // Collect instructions with negative FP constants from the subtree that ends 2114 // in Op. 2115 SmallVector<Instruction *, 4> Candidates; 2116 getNegatibleInsts(Op, Candidates); 2117 if (Candidates.empty()) 2118 return nullptr; 2119 2120 // Don't canonicalize x + (-Constant * y) -> x - (Constant * y), if the 2121 // resulting subtract will be broken up later. This can get us into an 2122 // infinite loop during reassociation. 2123 bool IsFSub = I->getOpcode() == Instruction::FSub; 2124 bool NeedsSubtract = !IsFSub && Candidates.size() % 2 == 1; 2125 if (NeedsSubtract && ShouldBreakUpSubtract(I)) 2126 return nullptr; 2127 2128 for (Instruction *Negatible : Candidates) { 2129 const APFloat *C; 2130 if (match(Negatible->getOperand(0), m_APFloat(C))) { 2131 assert(!match(Negatible->getOperand(1), m_Constant()) && 2132 "Expecting only 1 constant operand"); 2133 assert(C->isNegative() && "Expected negative FP constant"); 2134 Negatible->setOperand(0, ConstantFP::get(Negatible->getType(), abs(*C))); 2135 MadeChange = true; 2136 } 2137 if (match(Negatible->getOperand(1), m_APFloat(C))) { 2138 assert(!match(Negatible->getOperand(0), m_Constant()) && 2139 "Expecting only 1 constant operand"); 2140 assert(C->isNegative() && "Expected negative FP constant"); 2141 Negatible->setOperand(1, ConstantFP::get(Negatible->getType(), abs(*C))); 2142 MadeChange = true; 2143 } 2144 } 2145 assert(MadeChange == true && "Negative constant candidate was not changed"); 2146 2147 // Negations cancelled out. 2148 if (Candidates.size() % 2 == 0) 2149 return I; 2150 2151 // Negate the final operand in the expression by flipping the opcode of this 2152 // fadd/fsub. 2153 assert(Candidates.size() % 2 == 1 && "Expected odd number"); 2154 IRBuilder<> Builder(I); 2155 Value *NewInst = IsFSub ? Builder.CreateFAddFMF(OtherOp, Op, I) 2156 : Builder.CreateFSubFMF(OtherOp, Op, I); 2157 I->replaceAllUsesWith(NewInst); 2158 RedoInsts.insert(I); 2159 return dyn_cast<Instruction>(NewInst); 2160 } 2161 2162 /// Canonicalize expressions that contain a negative floating-point constant 2163 /// of the following form: 2164 /// OtherOp + (subtree) -> OtherOp {+/-} (canonical subtree) 2165 /// (subtree) + OtherOp -> OtherOp {+/-} (canonical subtree) 2166 /// OtherOp - (subtree) -> OtherOp {+/-} (canonical subtree) 2167 /// 2168 /// The fadd/fsub opcode may be switched to allow folding a negation into the 2169 /// input instruction. 2170 Instruction *ReassociatePass::canonicalizeNegFPConstants(Instruction *I) { 2171 LLVM_DEBUG(dbgs() << "Combine negations for: " << *I << '\n'); 2172 Value *X; 2173 Instruction *Op; 2174 if (match(I, m_FAdd(m_Value(X), m_OneUse(m_Instruction(Op))))) 2175 if (Instruction *R = canonicalizeNegFPConstantsForOp(I, Op, X)) 2176 I = R; 2177 if (match(I, m_FAdd(m_OneUse(m_Instruction(Op)), m_Value(X)))) 2178 if (Instruction *R = canonicalizeNegFPConstantsForOp(I, Op, X)) 2179 I = R; 2180 if (match(I, m_FSub(m_Value(X), m_OneUse(m_Instruction(Op))))) 2181 if (Instruction *R = canonicalizeNegFPConstantsForOp(I, Op, X)) 2182 I = R; 2183 return I; 2184 } 2185 2186 /// Inspect and optimize the given instruction. Note that erasing 2187 /// instructions is not allowed. 2188 void ReassociatePass::OptimizeInst(Instruction *I) { 2189 // Only consider operations that we understand. 2190 if (!isa<UnaryOperator>(I) && !isa<BinaryOperator>(I)) 2191 return; 2192 2193 if (I->getOpcode() == Instruction::Shl && isa<ConstantInt>(I->getOperand(1))) 2194 // If an operand of this shift is a reassociable multiply, or if the shift 2195 // is used by a reassociable multiply or add, turn into a multiply. 2196 if (isReassociableOp(I->getOperand(0), Instruction::Mul) || 2197 (I->hasOneUse() && 2198 (isReassociableOp(I->user_back(), Instruction::Mul) || 2199 isReassociableOp(I->user_back(), Instruction::Add)))) { 2200 Instruction *NI = ConvertShiftToMul(I); 2201 RedoInsts.insert(I); 2202 MadeChange = true; 2203 I = NI; 2204 } 2205 2206 // Commute binary operators, to canonicalize the order of their operands. 2207 // This can potentially expose more CSE opportunities, and makes writing other 2208 // transformations simpler. 2209 if (I->isCommutative()) 2210 canonicalizeOperands(I); 2211 2212 // Canonicalize negative constants out of expressions. 2213 if (Instruction *Res = canonicalizeNegFPConstants(I)) 2214 I = Res; 2215 2216 // Don't optimize floating-point instructions unless they have the 2217 // appropriate FastMathFlags for reassociation enabled. 2218 if (isa<FPMathOperator>(I) && !hasFPAssociativeFlags(I)) 2219 return; 2220 2221 // Do not reassociate boolean (i1) expressions. We want to preserve the 2222 // original order of evaluation for short-circuited comparisons that 2223 // SimplifyCFG has folded to AND/OR expressions. If the expression 2224 // is not further optimized, it is likely to be transformed back to a 2225 // short-circuited form for code gen, and the source order may have been 2226 // optimized for the most likely conditions. 2227 if (I->getType()->isIntegerTy(1)) 2228 return; 2229 2230 // If this is a bitwise or instruction of operands 2231 // with no common bits set, convert it to X+Y. 2232 if (I->getOpcode() == Instruction::Or && 2233 shouldConvertOrWithNoCommonBitsToAdd(I) && !isLoadCombineCandidate(I) && 2234 haveNoCommonBitsSet(I->getOperand(0), I->getOperand(1), 2235 I->getModule()->getDataLayout(), /*AC=*/nullptr, I, 2236 /*DT=*/nullptr)) { 2237 Instruction *NI = convertOrWithNoCommonBitsToAdd(I); 2238 RedoInsts.insert(I); 2239 MadeChange = true; 2240 I = NI; 2241 } 2242 2243 // If this is a subtract instruction which is not already in negate form, 2244 // see if we can convert it to X+-Y. 2245 if (I->getOpcode() == Instruction::Sub) { 2246 if (ShouldBreakUpSubtract(I)) { 2247 Instruction *NI = BreakUpSubtract(I, RedoInsts); 2248 RedoInsts.insert(I); 2249 MadeChange = true; 2250 I = NI; 2251 } else if (match(I, m_Neg(m_Value()))) { 2252 // Otherwise, this is a negation. See if the operand is a multiply tree 2253 // and if this is not an inner node of a multiply tree. 2254 if (isReassociableOp(I->getOperand(1), Instruction::Mul) && 2255 (!I->hasOneUse() || 2256 !isReassociableOp(I->user_back(), Instruction::Mul))) { 2257 Instruction *NI = LowerNegateToMultiply(I); 2258 // If the negate was simplified, revisit the users to see if we can 2259 // reassociate further. 2260 for (User *U : NI->users()) { 2261 if (BinaryOperator *Tmp = dyn_cast<BinaryOperator>(U)) 2262 RedoInsts.insert(Tmp); 2263 } 2264 RedoInsts.insert(I); 2265 MadeChange = true; 2266 I = NI; 2267 } 2268 } 2269 } else if (I->getOpcode() == Instruction::FNeg || 2270 I->getOpcode() == Instruction::FSub) { 2271 if (ShouldBreakUpSubtract(I)) { 2272 Instruction *NI = BreakUpSubtract(I, RedoInsts); 2273 RedoInsts.insert(I); 2274 MadeChange = true; 2275 I = NI; 2276 } else if (match(I, m_FNeg(m_Value()))) { 2277 // Otherwise, this is a negation. See if the operand is a multiply tree 2278 // and if this is not an inner node of a multiply tree. 2279 Value *Op = isa<BinaryOperator>(I) ? I->getOperand(1) : 2280 I->getOperand(0); 2281 if (isReassociableOp(Op, Instruction::FMul) && 2282 (!I->hasOneUse() || 2283 !isReassociableOp(I->user_back(), Instruction::FMul))) { 2284 // If the negate was simplified, revisit the users to see if we can 2285 // reassociate further. 2286 Instruction *NI = LowerNegateToMultiply(I); 2287 for (User *U : NI->users()) { 2288 if (BinaryOperator *Tmp = dyn_cast<BinaryOperator>(U)) 2289 RedoInsts.insert(Tmp); 2290 } 2291 RedoInsts.insert(I); 2292 MadeChange = true; 2293 I = NI; 2294 } 2295 } 2296 } 2297 2298 // If this instruction is an associative binary operator, process it. 2299 if (!I->isAssociative()) return; 2300 BinaryOperator *BO = cast<BinaryOperator>(I); 2301 2302 // If this is an interior node of a reassociable tree, ignore it until we 2303 // get to the root of the tree, to avoid N^2 analysis. 2304 unsigned Opcode = BO->getOpcode(); 2305 if (BO->hasOneUse() && BO->user_back()->getOpcode() == Opcode) { 2306 // During the initial run we will get to the root of the tree. 2307 // But if we get here while we are redoing instructions, there is no 2308 // guarantee that the root will be visited. So Redo later 2309 if (BO->user_back() != BO && 2310 BO->getParent() == BO->user_back()->getParent()) 2311 RedoInsts.insert(BO->user_back()); 2312 return; 2313 } 2314 2315 // If this is an add tree that is used by a sub instruction, ignore it 2316 // until we process the subtract. 2317 if (BO->hasOneUse() && BO->getOpcode() == Instruction::Add && 2318 cast<Instruction>(BO->user_back())->getOpcode() == Instruction::Sub) 2319 return; 2320 if (BO->hasOneUse() && BO->getOpcode() == Instruction::FAdd && 2321 cast<Instruction>(BO->user_back())->getOpcode() == Instruction::FSub) 2322 return; 2323 2324 ReassociateExpression(BO); 2325 } 2326 2327 void ReassociatePass::ReassociateExpression(BinaryOperator *I) { 2328 // First, walk the expression tree, linearizing the tree, collecting the 2329 // operand information. 2330 SmallVector<RepeatedValue, 8> Tree; 2331 MadeChange |= LinearizeExprTree(I, Tree, RedoInsts); 2332 SmallVector<ValueEntry, 8> Ops; 2333 Ops.reserve(Tree.size()); 2334 for (const RepeatedValue &E : Tree) 2335 Ops.append(E.second.getZExtValue(), ValueEntry(getRank(E.first), E.first)); 2336 2337 LLVM_DEBUG(dbgs() << "RAIn:\t"; PrintOps(I, Ops); dbgs() << '\n'); 2338 2339 // Now that we have linearized the tree to a list and have gathered all of 2340 // the operands and their ranks, sort the operands by their rank. Use a 2341 // stable_sort so that values with equal ranks will have their relative 2342 // positions maintained (and so the compiler is deterministic). Note that 2343 // this sorts so that the highest ranking values end up at the beginning of 2344 // the vector. 2345 llvm::stable_sort(Ops); 2346 2347 // Now that we have the expression tree in a convenient 2348 // sorted form, optimize it globally if possible. 2349 if (Value *V = OptimizeExpression(I, Ops)) { 2350 if (V == I) 2351 // Self-referential expression in unreachable code. 2352 return; 2353 // This expression tree simplified to something that isn't a tree, 2354 // eliminate it. 2355 LLVM_DEBUG(dbgs() << "Reassoc to scalar: " << *V << '\n'); 2356 I->replaceAllUsesWith(V); 2357 if (Instruction *VI = dyn_cast<Instruction>(V)) 2358 if (I->getDebugLoc()) 2359 VI->setDebugLoc(I->getDebugLoc()); 2360 RedoInsts.insert(I); 2361 ++NumAnnihil; 2362 return; 2363 } 2364 2365 // We want to sink immediates as deeply as possible except in the case where 2366 // this is a multiply tree used only by an add, and the immediate is a -1. 2367 // In this case we reassociate to put the negation on the outside so that we 2368 // can fold the negation into the add: (-X)*Y + Z -> Z-X*Y 2369 if (I->hasOneUse()) { 2370 if (I->getOpcode() == Instruction::Mul && 2371 cast<Instruction>(I->user_back())->getOpcode() == Instruction::Add && 2372 isa<ConstantInt>(Ops.back().Op) && 2373 cast<ConstantInt>(Ops.back().Op)->isMinusOne()) { 2374 ValueEntry Tmp = Ops.pop_back_val(); 2375 Ops.insert(Ops.begin(), Tmp); 2376 } else if (I->getOpcode() == Instruction::FMul && 2377 cast<Instruction>(I->user_back())->getOpcode() == 2378 Instruction::FAdd && 2379 isa<ConstantFP>(Ops.back().Op) && 2380 cast<ConstantFP>(Ops.back().Op)->isExactlyValue(-1.0)) { 2381 ValueEntry Tmp = Ops.pop_back_val(); 2382 Ops.insert(Ops.begin(), Tmp); 2383 } 2384 } 2385 2386 LLVM_DEBUG(dbgs() << "RAOut:\t"; PrintOps(I, Ops); dbgs() << '\n'); 2387 2388 if (Ops.size() == 1) { 2389 if (Ops[0].Op == I) 2390 // Self-referential expression in unreachable code. 2391 return; 2392 2393 // This expression tree simplified to something that isn't a tree, 2394 // eliminate it. 2395 I->replaceAllUsesWith(Ops[0].Op); 2396 if (Instruction *OI = dyn_cast<Instruction>(Ops[0].Op)) 2397 OI->setDebugLoc(I->getDebugLoc()); 2398 RedoInsts.insert(I); 2399 return; 2400 } 2401 2402 if (Ops.size() > 2 && Ops.size() <= GlobalReassociateLimit) { 2403 // Find the pair with the highest count in the pairmap and move it to the 2404 // back of the list so that it can later be CSE'd. 2405 // example: 2406 // a*b*c*d*e 2407 // if c*e is the most "popular" pair, we can express this as 2408 // (((c*e)*d)*b)*a 2409 unsigned Max = 1; 2410 unsigned BestRank = 0; 2411 std::pair<unsigned, unsigned> BestPair; 2412 unsigned Idx = I->getOpcode() - Instruction::BinaryOpsBegin; 2413 for (unsigned i = 0; i < Ops.size() - 1; ++i) 2414 for (unsigned j = i + 1; j < Ops.size(); ++j) { 2415 unsigned Score = 0; 2416 Value *Op0 = Ops[i].Op; 2417 Value *Op1 = Ops[j].Op; 2418 if (std::less<Value *>()(Op1, Op0)) 2419 std::swap(Op0, Op1); 2420 auto it = PairMap[Idx].find({Op0, Op1}); 2421 if (it != PairMap[Idx].end()) { 2422 // Functions like BreakUpSubtract() can erase the Values we're using 2423 // as keys and create new Values after we built the PairMap. There's a 2424 // small chance that the new nodes can have the same address as 2425 // something already in the table. We shouldn't accumulate the stored 2426 // score in that case as it refers to the wrong Value. 2427 if (it->second.isValid()) 2428 Score += it->second.Score; 2429 } 2430 2431 unsigned MaxRank = std::max(Ops[i].Rank, Ops[j].Rank); 2432 if (Score > Max || (Score == Max && MaxRank < BestRank)) { 2433 BestPair = {i, j}; 2434 Max = Score; 2435 BestRank = MaxRank; 2436 } 2437 } 2438 if (Max > 1) { 2439 auto Op0 = Ops[BestPair.first]; 2440 auto Op1 = Ops[BestPair.second]; 2441 Ops.erase(&Ops[BestPair.second]); 2442 Ops.erase(&Ops[BestPair.first]); 2443 Ops.push_back(Op0); 2444 Ops.push_back(Op1); 2445 } 2446 } 2447 // Now that we ordered and optimized the expressions, splat them back into 2448 // the expression tree, removing any unneeded nodes. 2449 RewriteExprTree(I, Ops); 2450 } 2451 2452 void 2453 ReassociatePass::BuildPairMap(ReversePostOrderTraversal<Function *> &RPOT) { 2454 // Make a "pairmap" of how often each operand pair occurs. 2455 for (BasicBlock *BI : RPOT) { 2456 for (Instruction &I : *BI) { 2457 if (!I.isAssociative()) 2458 continue; 2459 2460 // Ignore nodes that aren't at the root of trees. 2461 if (I.hasOneUse() && I.user_back()->getOpcode() == I.getOpcode()) 2462 continue; 2463 2464 // Collect all operands in a single reassociable expression. 2465 // Since Reassociate has already been run once, we can assume things 2466 // are already canonical according to Reassociation's regime. 2467 SmallVector<Value *, 8> Worklist = { I.getOperand(0), I.getOperand(1) }; 2468 SmallVector<Value *, 8> Ops; 2469 while (!Worklist.empty() && Ops.size() <= GlobalReassociateLimit) { 2470 Value *Op = Worklist.pop_back_val(); 2471 Instruction *OpI = dyn_cast<Instruction>(Op); 2472 if (!OpI || OpI->getOpcode() != I.getOpcode() || !OpI->hasOneUse()) { 2473 Ops.push_back(Op); 2474 continue; 2475 } 2476 // Be paranoid about self-referencing expressions in unreachable code. 2477 if (OpI->getOperand(0) != OpI) 2478 Worklist.push_back(OpI->getOperand(0)); 2479 if (OpI->getOperand(1) != OpI) 2480 Worklist.push_back(OpI->getOperand(1)); 2481 } 2482 // Skip extremely long expressions. 2483 if (Ops.size() > GlobalReassociateLimit) 2484 continue; 2485 2486 // Add all pairwise combinations of operands to the pair map. 2487 unsigned BinaryIdx = I.getOpcode() - Instruction::BinaryOpsBegin; 2488 SmallSet<std::pair<Value *, Value*>, 32> Visited; 2489 for (unsigned i = 0; i < Ops.size() - 1; ++i) { 2490 for (unsigned j = i + 1; j < Ops.size(); ++j) { 2491 // Canonicalize operand orderings. 2492 Value *Op0 = Ops[i]; 2493 Value *Op1 = Ops[j]; 2494 if (std::less<Value *>()(Op1, Op0)) 2495 std::swap(Op0, Op1); 2496 if (!Visited.insert({Op0, Op1}).second) 2497 continue; 2498 auto res = PairMap[BinaryIdx].insert({{Op0, Op1}, {Op0, Op1, 1}}); 2499 if (!res.second) { 2500 // If either key value has been erased then we've got the same 2501 // address by coincidence. That can't happen here because nothing is 2502 // erasing values but it can happen by the time we're querying the 2503 // map. 2504 assert(res.first->second.isValid() && "WeakVH invalidated"); 2505 ++res.first->second.Score; 2506 } 2507 } 2508 } 2509 } 2510 } 2511 } 2512 2513 PreservedAnalyses ReassociatePass::run(Function &F, FunctionAnalysisManager &) { 2514 // Get the functions basic blocks in Reverse Post Order. This order is used by 2515 // BuildRankMap to pre calculate ranks correctly. It also excludes dead basic 2516 // blocks (it has been seen that the analysis in this pass could hang when 2517 // analysing dead basic blocks). 2518 ReversePostOrderTraversal<Function *> RPOT(&F); 2519 2520 // Calculate the rank map for F. 2521 BuildRankMap(F, RPOT); 2522 2523 // Build the pair map before running reassociate. 2524 // Technically this would be more accurate if we did it after one round 2525 // of reassociation, but in practice it doesn't seem to help much on 2526 // real-world code, so don't waste the compile time running reassociate 2527 // twice. 2528 // If a user wants, they could expicitly run reassociate twice in their 2529 // pass pipeline for further potential gains. 2530 // It might also be possible to update the pair map during runtime, but the 2531 // overhead of that may be large if there's many reassociable chains. 2532 BuildPairMap(RPOT); 2533 2534 MadeChange = false; 2535 2536 // Traverse the same blocks that were analysed by BuildRankMap. 2537 for (BasicBlock *BI : RPOT) { 2538 assert(RankMap.count(&*BI) && "BB should be ranked."); 2539 // Optimize every instruction in the basic block. 2540 for (BasicBlock::iterator II = BI->begin(), IE = BI->end(); II != IE;) 2541 if (isInstructionTriviallyDead(&*II)) { 2542 EraseInst(&*II++); 2543 } else { 2544 OptimizeInst(&*II); 2545 assert(II->getParent() == &*BI && "Moved to a different block!"); 2546 ++II; 2547 } 2548 2549 // Make a copy of all the instructions to be redone so we can remove dead 2550 // instructions. 2551 OrderedSet ToRedo(RedoInsts); 2552 // Iterate over all instructions to be reevaluated and remove trivially dead 2553 // instructions. If any operand of the trivially dead instruction becomes 2554 // dead mark it for deletion as well. Continue this process until all 2555 // trivially dead instructions have been removed. 2556 while (!ToRedo.empty()) { 2557 Instruction *I = ToRedo.pop_back_val(); 2558 if (isInstructionTriviallyDead(I)) { 2559 RecursivelyEraseDeadInsts(I, ToRedo); 2560 MadeChange = true; 2561 } 2562 } 2563 2564 // Now that we have removed dead instructions, we can reoptimize the 2565 // remaining instructions. 2566 while (!RedoInsts.empty()) { 2567 Instruction *I = RedoInsts.front(); 2568 RedoInsts.erase(RedoInsts.begin()); 2569 if (isInstructionTriviallyDead(I)) 2570 EraseInst(I); 2571 else 2572 OptimizeInst(I); 2573 } 2574 } 2575 2576 // We are done with the rank map and pair map. 2577 RankMap.clear(); 2578 ValueRankMap.clear(); 2579 for (auto &Entry : PairMap) 2580 Entry.clear(); 2581 2582 if (MadeChange) { 2583 PreservedAnalyses PA; 2584 PA.preserveSet<CFGAnalyses>(); 2585 return PA; 2586 } 2587 2588 return PreservedAnalyses::all(); 2589 } 2590 2591 namespace { 2592 2593 class ReassociateLegacyPass : public FunctionPass { 2594 ReassociatePass Impl; 2595 2596 public: 2597 static char ID; // Pass identification, replacement for typeid 2598 2599 ReassociateLegacyPass() : FunctionPass(ID) { 2600 initializeReassociateLegacyPassPass(*PassRegistry::getPassRegistry()); 2601 } 2602 2603 bool runOnFunction(Function &F) override { 2604 if (skipFunction(F)) 2605 return false; 2606 2607 FunctionAnalysisManager DummyFAM; 2608 auto PA = Impl.run(F, DummyFAM); 2609 return !PA.areAllPreserved(); 2610 } 2611 2612 void getAnalysisUsage(AnalysisUsage &AU) const override { 2613 AU.setPreservesCFG(); 2614 AU.addPreserved<AAResultsWrapperPass>(); 2615 AU.addPreserved<BasicAAWrapperPass>(); 2616 AU.addPreserved<GlobalsAAWrapperPass>(); 2617 } 2618 }; 2619 2620 } // end anonymous namespace 2621 2622 char ReassociateLegacyPass::ID = 0; 2623 2624 INITIALIZE_PASS(ReassociateLegacyPass, "reassociate", 2625 "Reassociate expressions", false, false) 2626 2627 // Public interface to the Reassociate pass 2628 FunctionPass *llvm::createReassociatePass() { 2629 return new ReassociateLegacyPass(); 2630 } 2631