1 //===- InstCombineAndOrXor.cpp --------------------------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements the visitAnd, visitOr, and visitXor functions.
12 //===----------------------------------------------------------------------===//
14 #include "InstCombineInternal.h"
15 #include "llvm/Analysis/InstructionSimplify.h"
16 #include "llvm/IR/ConstantRange.h"
17 #include "llvm/IR/Intrinsics.h"
18 #include "llvm/IR/PatternMatch.h"
19 #include "llvm/Transforms/Utils/CmpInstAnalysis.h"
21 using namespace PatternMatch;
23 #define DEBUG_TYPE "instcombine"
25 static inline Value *dyn_castNotVal(Value *V) {
26 // If this is not(not(x)) don't return that this is a not: we want the two
27 // not's to be folded first.
28 if (BinaryOperator::isNot(V)) {
29 Value *Operand = BinaryOperator::getNotArgument(V);
30 if (!IsFreeToInvert(Operand, Operand->hasOneUse()))
34 // Constants can be considered to be not'ed values...
35 if (ConstantInt *C = dyn_cast<ConstantInt>(V))
36 return ConstantInt::get(C->getType(), ~C->getValue());
40 /// getFCmpCode - Similar to getICmpCode but for FCmpInst. This encodes a fcmp
41 /// predicate into a three bit mask. It also returns whether it is an ordered
42 /// predicate by reference.
43 static unsigned getFCmpCode(FCmpInst::Predicate CC, bool &isOrdered) {
46 case FCmpInst::FCMP_ORD: isOrdered = true; return 0; // 000
47 case FCmpInst::FCMP_UNO: return 0; // 000
48 case FCmpInst::FCMP_OGT: isOrdered = true; return 1; // 001
49 case FCmpInst::FCMP_UGT: return 1; // 001
50 case FCmpInst::FCMP_OEQ: isOrdered = true; return 2; // 010
51 case FCmpInst::FCMP_UEQ: return 2; // 010
52 case FCmpInst::FCMP_OGE: isOrdered = true; return 3; // 011
53 case FCmpInst::FCMP_UGE: return 3; // 011
54 case FCmpInst::FCMP_OLT: isOrdered = true; return 4; // 100
55 case FCmpInst::FCMP_ULT: return 4; // 100
56 case FCmpInst::FCMP_ONE: isOrdered = true; return 5; // 101
57 case FCmpInst::FCMP_UNE: return 5; // 101
58 case FCmpInst::FCMP_OLE: isOrdered = true; return 6; // 110
59 case FCmpInst::FCMP_ULE: return 6; // 110
62 // Not expecting FCMP_FALSE and FCMP_TRUE;
63 llvm_unreachable("Unexpected FCmp predicate!");
67 /// getNewICmpValue - This is the complement of getICmpCode, which turns an
68 /// opcode and two operands into either a constant true or false, or a brand
69 /// new ICmp instruction. The sign is passed in to determine which kind
70 /// of predicate to use in the new icmp instruction.
71 static Value *getNewICmpValue(bool Sign, unsigned Code, Value *LHS, Value *RHS,
72 InstCombiner::BuilderTy *Builder) {
73 ICmpInst::Predicate NewPred;
74 if (Value *NewConstant = getICmpValue(Sign, Code, LHS, RHS, NewPred))
76 return Builder->CreateICmp(NewPred, LHS, RHS);
79 /// getFCmpValue - This is the complement of getFCmpCode, which turns an
80 /// opcode and two operands into either a FCmp instruction. isordered is passed
81 /// in to determine which kind of predicate to use in the new fcmp instruction.
82 static Value *getFCmpValue(bool isordered, unsigned code,
83 Value *LHS, Value *RHS,
84 InstCombiner::BuilderTy *Builder) {
85 CmpInst::Predicate Pred;
87 default: llvm_unreachable("Illegal FCmp code!");
88 case 0: Pred = isordered ? FCmpInst::FCMP_ORD : FCmpInst::FCMP_UNO; break;
89 case 1: Pred = isordered ? FCmpInst::FCMP_OGT : FCmpInst::FCMP_UGT; break;
90 case 2: Pred = isordered ? FCmpInst::FCMP_OEQ : FCmpInst::FCMP_UEQ; break;
91 case 3: Pred = isordered ? FCmpInst::FCMP_OGE : FCmpInst::FCMP_UGE; break;
92 case 4: Pred = isordered ? FCmpInst::FCMP_OLT : FCmpInst::FCMP_ULT; break;
93 case 5: Pred = isordered ? FCmpInst::FCMP_ONE : FCmpInst::FCMP_UNE; break;
94 case 6: Pred = isordered ? FCmpInst::FCMP_OLE : FCmpInst::FCMP_ULE; break;
96 if (!isordered) return ConstantInt::getTrue(LHS->getContext());
97 Pred = FCmpInst::FCMP_ORD; break;
99 return Builder->CreateFCmp(Pred, LHS, RHS);
102 /// \brief Transform BITWISE_OP(BSWAP(A),BSWAP(B)) to BSWAP(BITWISE_OP(A, B))
103 /// \param I Binary operator to transform.
104 /// \return Pointer to node that must replace the original binary operator, or
105 /// null pointer if no transformation was made.
106 Value *InstCombiner::SimplifyBSwap(BinaryOperator &I) {
107 IntegerType *ITy = dyn_cast<IntegerType>(I.getType());
110 if (I.getType()->isVectorTy()) return nullptr;
112 // Can only do bitwise ops.
113 unsigned Op = I.getOpcode();
114 if (Op != Instruction::And && Op != Instruction::Or &&
115 Op != Instruction::Xor)
118 Value *OldLHS = I.getOperand(0);
119 Value *OldRHS = I.getOperand(1);
120 ConstantInt *ConstLHS = dyn_cast<ConstantInt>(OldLHS);
121 ConstantInt *ConstRHS = dyn_cast<ConstantInt>(OldRHS);
122 IntrinsicInst *IntrLHS = dyn_cast<IntrinsicInst>(OldLHS);
123 IntrinsicInst *IntrRHS = dyn_cast<IntrinsicInst>(OldRHS);
124 bool IsBswapLHS = (IntrLHS && IntrLHS->getIntrinsicID() == Intrinsic::bswap);
125 bool IsBswapRHS = (IntrRHS && IntrRHS->getIntrinsicID() == Intrinsic::bswap);
127 if (!IsBswapLHS && !IsBswapRHS)
130 if (!IsBswapLHS && !ConstLHS)
133 if (!IsBswapRHS && !ConstRHS)
136 /// OP( BSWAP(x), BSWAP(y) ) -> BSWAP( OP(x, y) )
137 /// OP( BSWAP(x), CONSTANT ) -> BSWAP( OP(x, BSWAP(CONSTANT) ) )
138 Value *NewLHS = IsBswapLHS ? IntrLHS->getOperand(0) :
139 Builder->getInt(ConstLHS->getValue().byteSwap());
141 Value *NewRHS = IsBswapRHS ? IntrRHS->getOperand(0) :
142 Builder->getInt(ConstRHS->getValue().byteSwap());
144 Value *BinOp = nullptr;
145 if (Op == Instruction::And)
146 BinOp = Builder->CreateAnd(NewLHS, NewRHS);
147 else if (Op == Instruction::Or)
148 BinOp = Builder->CreateOr(NewLHS, NewRHS);
149 else //if (Op == Instruction::Xor)
150 BinOp = Builder->CreateXor(NewLHS, NewRHS);
152 Module *M = I.getParent()->getParent()->getParent();
153 Function *F = Intrinsic::getDeclaration(M, Intrinsic::bswap, ITy);
154 return Builder->CreateCall(F, BinOp);
157 // OptAndOp - This handles expressions of the form ((val OP C1) & C2). Where
158 // the Op parameter is 'OP', OpRHS is 'C1', and AndRHS is 'C2'. Op is
159 // guaranteed to be a binary operator.
160 Instruction *InstCombiner::OptAndOp(Instruction *Op,
163 BinaryOperator &TheAnd) {
164 Value *X = Op->getOperand(0);
165 Constant *Together = nullptr;
167 Together = ConstantExpr::getAnd(AndRHS, OpRHS);
169 switch (Op->getOpcode()) {
170 case Instruction::Xor:
171 if (Op->hasOneUse()) {
172 // (X ^ C1) & C2 --> (X & C2) ^ (C1&C2)
173 Value *And = Builder->CreateAnd(X, AndRHS);
175 return BinaryOperator::CreateXor(And, Together);
178 case Instruction::Or:
179 if (Op->hasOneUse()){
180 if (Together != OpRHS) {
181 // (X | C1) & C2 --> (X | (C1&C2)) & C2
182 Value *Or = Builder->CreateOr(X, Together);
184 return BinaryOperator::CreateAnd(Or, AndRHS);
187 ConstantInt *TogetherCI = dyn_cast<ConstantInt>(Together);
188 if (TogetherCI && !TogetherCI->isZero()){
189 // (X | C1) & C2 --> (X & (C2^(C1&C2))) | C1
190 // NOTE: This reduces the number of bits set in the & mask, which
191 // can expose opportunities for store narrowing.
192 Together = ConstantExpr::getXor(AndRHS, Together);
193 Value *And = Builder->CreateAnd(X, Together);
195 return BinaryOperator::CreateOr(And, OpRHS);
200 case Instruction::Add:
201 if (Op->hasOneUse()) {
202 // Adding a one to a single bit bit-field should be turned into an XOR
203 // of the bit. First thing to check is to see if this AND is with a
204 // single bit constant.
205 const APInt &AndRHSV = AndRHS->getValue();
207 // If there is only one bit set.
208 if (AndRHSV.isPowerOf2()) {
209 // Ok, at this point, we know that we are masking the result of the
210 // ADD down to exactly one bit. If the constant we are adding has
211 // no bits set below this bit, then we can eliminate the ADD.
212 const APInt& AddRHS = OpRHS->getValue();
214 // Check to see if any bits below the one bit set in AndRHSV are set.
215 if ((AddRHS & (AndRHSV-1)) == 0) {
216 // If not, the only thing that can effect the output of the AND is
217 // the bit specified by AndRHSV. If that bit is set, the effect of
218 // the XOR is to toggle the bit. If it is clear, then the ADD has
220 if ((AddRHS & AndRHSV) == 0) { // Bit is not set, noop
221 TheAnd.setOperand(0, X);
224 // Pull the XOR out of the AND.
225 Value *NewAnd = Builder->CreateAnd(X, AndRHS);
226 NewAnd->takeName(Op);
227 return BinaryOperator::CreateXor(NewAnd, AndRHS);
234 case Instruction::Shl: {
235 // We know that the AND will not produce any of the bits shifted in, so if
236 // the anded constant includes them, clear them now!
238 uint32_t BitWidth = AndRHS->getType()->getBitWidth();
239 uint32_t OpRHSVal = OpRHS->getLimitedValue(BitWidth);
240 APInt ShlMask(APInt::getHighBitsSet(BitWidth, BitWidth-OpRHSVal));
241 ConstantInt *CI = Builder->getInt(AndRHS->getValue() & ShlMask);
243 if (CI->getValue() == ShlMask)
244 // Masking out bits that the shift already masks.
245 return ReplaceInstUsesWith(TheAnd, Op); // No need for the and.
247 if (CI != AndRHS) { // Reducing bits set in and.
248 TheAnd.setOperand(1, CI);
253 case Instruction::LShr: {
254 // We know that the AND will not produce any of the bits shifted in, so if
255 // the anded constant includes them, clear them now! This only applies to
256 // unsigned shifts, because a signed shr may bring in set bits!
258 uint32_t BitWidth = AndRHS->getType()->getBitWidth();
259 uint32_t OpRHSVal = OpRHS->getLimitedValue(BitWidth);
260 APInt ShrMask(APInt::getLowBitsSet(BitWidth, BitWidth - OpRHSVal));
261 ConstantInt *CI = Builder->getInt(AndRHS->getValue() & ShrMask);
263 if (CI->getValue() == ShrMask)
264 // Masking out bits that the shift already masks.
265 return ReplaceInstUsesWith(TheAnd, Op);
268 TheAnd.setOperand(1, CI); // Reduce bits set in and cst.
273 case Instruction::AShr:
275 // See if this is shifting in some sign extension, then masking it out
277 if (Op->hasOneUse()) {
278 uint32_t BitWidth = AndRHS->getType()->getBitWidth();
279 uint32_t OpRHSVal = OpRHS->getLimitedValue(BitWidth);
280 APInt ShrMask(APInt::getLowBitsSet(BitWidth, BitWidth - OpRHSVal));
281 Constant *C = Builder->getInt(AndRHS->getValue() & ShrMask);
282 if (C == AndRHS) { // Masking out bits shifted in.
283 // (Val ashr C1) & C2 -> (Val lshr C1) & C2
284 // Make the argument unsigned.
285 Value *ShVal = Op->getOperand(0);
286 ShVal = Builder->CreateLShr(ShVal, OpRHS, Op->getName());
287 return BinaryOperator::CreateAnd(ShVal, AndRHS, TheAnd.getName());
295 /// Emit a computation of: (V >= Lo && V < Hi) if Inside is true, otherwise
296 /// (V < Lo || V >= Hi). In practice, we emit the more efficient
297 /// (V-Lo) \<u Hi-Lo. This method expects that Lo <= Hi. isSigned indicates
298 /// whether to treat the V, Lo and HI as signed or not. IB is the location to
299 /// insert new instructions.
300 Value *InstCombiner::InsertRangeTest(Value *V, Constant *Lo, Constant *Hi,
301 bool isSigned, bool Inside) {
302 assert(cast<ConstantInt>(ConstantExpr::getICmp((isSigned ?
303 ICmpInst::ICMP_SLE:ICmpInst::ICMP_ULE), Lo, Hi))->getZExtValue() &&
304 "Lo is not <= Hi in range emission code!");
307 if (Lo == Hi) // Trivially false.
308 return Builder->getFalse();
310 // V >= Min && V < Hi --> V < Hi
311 if (cast<ConstantInt>(Lo)->isMinValue(isSigned)) {
312 ICmpInst::Predicate pred = (isSigned ?
313 ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT);
314 return Builder->CreateICmp(pred, V, Hi);
317 // Emit V-Lo <u Hi-Lo
318 Constant *NegLo = ConstantExpr::getNeg(Lo);
319 Value *Add = Builder->CreateAdd(V, NegLo, V->getName()+".off");
320 Constant *UpperBound = ConstantExpr::getAdd(NegLo, Hi);
321 return Builder->CreateICmpULT(Add, UpperBound);
324 if (Lo == Hi) // Trivially true.
325 return Builder->getTrue();
327 // V < Min || V >= Hi -> V > Hi-1
328 Hi = SubOne(cast<ConstantInt>(Hi));
329 if (cast<ConstantInt>(Lo)->isMinValue(isSigned)) {
330 ICmpInst::Predicate pred = (isSigned ?
331 ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT);
332 return Builder->CreateICmp(pred, V, Hi);
335 // Emit V-Lo >u Hi-1-Lo
336 // Note that Hi has already had one subtracted from it, above.
337 ConstantInt *NegLo = cast<ConstantInt>(ConstantExpr::getNeg(Lo));
338 Value *Add = Builder->CreateAdd(V, NegLo, V->getName()+".off");
339 Constant *LowerBound = ConstantExpr::getAdd(NegLo, Hi);
340 return Builder->CreateICmpUGT(Add, LowerBound);
343 // isRunOfOnes - Returns true iff Val consists of one contiguous run of 1s with
344 // any number of 0s on either side. The 1s are allowed to wrap from LSB to
345 // MSB, so 0x000FFF0, 0x0000FFFF, and 0xFF0000FF are all runs. 0x0F0F0000 is
346 // not, since all 1s are not contiguous.
347 static bool isRunOfOnes(ConstantInt *Val, uint32_t &MB, uint32_t &ME) {
348 const APInt& V = Val->getValue();
349 uint32_t BitWidth = Val->getType()->getBitWidth();
350 if (!APIntOps::isShiftedMask(BitWidth, V)) return false;
352 // look for the first zero bit after the run of ones
353 MB = BitWidth - ((V - 1) ^ V).countLeadingZeros();
354 // look for the first non-zero bit
355 ME = V.getActiveBits();
359 /// FoldLogicalPlusAnd - This is part of an expression (LHS +/- RHS) & Mask,
360 /// where isSub determines whether the operator is a sub. If we can fold one of
361 /// the following xforms:
363 /// ((A & N) +/- B) & Mask -> (A +/- B) & Mask iff N&Mask == Mask
364 /// ((A | N) +/- B) & Mask -> (A +/- B) & Mask iff N&Mask == 0
365 /// ((A ^ N) +/- B) & Mask -> (A +/- B) & Mask iff N&Mask == 0
367 /// return (A +/- B).
369 Value *InstCombiner::FoldLogicalPlusAnd(Value *LHS, Value *RHS,
370 ConstantInt *Mask, bool isSub,
372 Instruction *LHSI = dyn_cast<Instruction>(LHS);
373 if (!LHSI || LHSI->getNumOperands() != 2 ||
374 !isa<ConstantInt>(LHSI->getOperand(1))) return nullptr;
376 ConstantInt *N = cast<ConstantInt>(LHSI->getOperand(1));
378 switch (LHSI->getOpcode()) {
379 default: return nullptr;
380 case Instruction::And:
381 if (ConstantExpr::getAnd(N, Mask) == Mask) {
382 // If the AndRHS is a power of two minus one (0+1+), this is simple.
383 if ((Mask->getValue().countLeadingZeros() +
384 Mask->getValue().countPopulation()) ==
385 Mask->getValue().getBitWidth())
388 // Otherwise, if Mask is 0+1+0+, and if B is known to have the low 0+
389 // part, we don't need any explicit masks to take them out of A. If that
390 // is all N is, ignore it.
391 uint32_t MB = 0, ME = 0;
392 if (isRunOfOnes(Mask, MB, ME)) { // begin/end bit of run, inclusive
393 uint32_t BitWidth = cast<IntegerType>(RHS->getType())->getBitWidth();
394 APInt Mask(APInt::getLowBitsSet(BitWidth, MB-1));
395 if (MaskedValueIsZero(RHS, Mask, 0, &I))
400 case Instruction::Or:
401 case Instruction::Xor:
402 // If the AndRHS is a power of two minus one (0+1+), and N&Mask == 0
403 if ((Mask->getValue().countLeadingZeros() +
404 Mask->getValue().countPopulation()) == Mask->getValue().getBitWidth()
405 && ConstantExpr::getAnd(N, Mask)->isNullValue())
411 return Builder->CreateSub(LHSI->getOperand(0), RHS, "fold");
412 return Builder->CreateAdd(LHSI->getOperand(0), RHS, "fold");
415 /// enum for classifying (icmp eq (A & B), C) and (icmp ne (A & B), C)
416 /// One of A and B is considered the mask, the other the value. This is
417 /// described as the "AMask" or "BMask" part of the enum. If the enum
418 /// contains only "Mask", then both A and B can be considered masks.
419 /// If A is the mask, then it was proven, that (A & C) == C. This
420 /// is trivial if C == A, or C == 0. If both A and C are constants, this
421 /// proof is also easy.
422 /// For the following explanations we assume that A is the mask.
423 /// The part "AllOnes" declares, that the comparison is true only
424 /// if (A & B) == A, or all bits of A are set in B.
425 /// Example: (icmp eq (A & 3), 3) -> FoldMskICmp_AMask_AllOnes
426 /// The part "AllZeroes" declares, that the comparison is true only
427 /// if (A & B) == 0, or all bits of A are cleared in B.
428 /// Example: (icmp eq (A & 3), 0) -> FoldMskICmp_Mask_AllZeroes
429 /// The part "Mixed" declares, that (A & B) == C and C might or might not
430 /// contain any number of one bits and zero bits.
431 /// Example: (icmp eq (A & 3), 1) -> FoldMskICmp_AMask_Mixed
432 /// The Part "Not" means, that in above descriptions "==" should be replaced
434 /// Example: (icmp ne (A & 3), 3) -> FoldMskICmp_AMask_NotAllOnes
435 /// If the mask A contains a single bit, then the following is equivalent:
436 /// (icmp eq (A & B), A) equals (icmp ne (A & B), 0)
437 /// (icmp ne (A & B), A) equals (icmp eq (A & B), 0)
438 enum MaskedICmpType {
439 FoldMskICmp_AMask_AllOnes = 1,
440 FoldMskICmp_AMask_NotAllOnes = 2,
441 FoldMskICmp_BMask_AllOnes = 4,
442 FoldMskICmp_BMask_NotAllOnes = 8,
443 FoldMskICmp_Mask_AllZeroes = 16,
444 FoldMskICmp_Mask_NotAllZeroes = 32,
445 FoldMskICmp_AMask_Mixed = 64,
446 FoldMskICmp_AMask_NotMixed = 128,
447 FoldMskICmp_BMask_Mixed = 256,
448 FoldMskICmp_BMask_NotMixed = 512
451 /// return the set of pattern classes (from MaskedICmpType)
452 /// that (icmp SCC (A & B), C) satisfies
453 static unsigned getTypeOfMaskedICmp(Value* A, Value* B, Value* C,
454 ICmpInst::Predicate SCC)
456 ConstantInt *ACst = dyn_cast<ConstantInt>(A);
457 ConstantInt *BCst = dyn_cast<ConstantInt>(B);
458 ConstantInt *CCst = dyn_cast<ConstantInt>(C);
459 bool icmp_eq = (SCC == ICmpInst::ICMP_EQ);
460 bool icmp_abit = (ACst && !ACst->isZero() &&
461 ACst->getValue().isPowerOf2());
462 bool icmp_bbit = (BCst && !BCst->isZero() &&
463 BCst->getValue().isPowerOf2());
465 if (CCst && CCst->isZero()) {
466 // if C is zero, then both A and B qualify as mask
467 result |= (icmp_eq ? (FoldMskICmp_Mask_AllZeroes |
468 FoldMskICmp_Mask_AllZeroes |
469 FoldMskICmp_AMask_Mixed |
470 FoldMskICmp_BMask_Mixed)
471 : (FoldMskICmp_Mask_NotAllZeroes |
472 FoldMskICmp_Mask_NotAllZeroes |
473 FoldMskICmp_AMask_NotMixed |
474 FoldMskICmp_BMask_NotMixed));
476 result |= (icmp_eq ? (FoldMskICmp_AMask_NotAllOnes |
477 FoldMskICmp_AMask_NotMixed)
478 : (FoldMskICmp_AMask_AllOnes |
479 FoldMskICmp_AMask_Mixed));
481 result |= (icmp_eq ? (FoldMskICmp_BMask_NotAllOnes |
482 FoldMskICmp_BMask_NotMixed)
483 : (FoldMskICmp_BMask_AllOnes |
484 FoldMskICmp_BMask_Mixed));
488 result |= (icmp_eq ? (FoldMskICmp_AMask_AllOnes |
489 FoldMskICmp_AMask_Mixed)
490 : (FoldMskICmp_AMask_NotAllOnes |
491 FoldMskICmp_AMask_NotMixed));
493 result |= (icmp_eq ? (FoldMskICmp_Mask_NotAllZeroes |
494 FoldMskICmp_AMask_NotMixed)
495 : (FoldMskICmp_Mask_AllZeroes |
496 FoldMskICmp_AMask_Mixed));
497 } else if (ACst && CCst &&
498 ConstantExpr::getAnd(ACst, CCst) == CCst) {
499 result |= (icmp_eq ? FoldMskICmp_AMask_Mixed
500 : FoldMskICmp_AMask_NotMixed);
503 result |= (icmp_eq ? (FoldMskICmp_BMask_AllOnes |
504 FoldMskICmp_BMask_Mixed)
505 : (FoldMskICmp_BMask_NotAllOnes |
506 FoldMskICmp_BMask_NotMixed));
508 result |= (icmp_eq ? (FoldMskICmp_Mask_NotAllZeroes |
509 FoldMskICmp_BMask_NotMixed)
510 : (FoldMskICmp_Mask_AllZeroes |
511 FoldMskICmp_BMask_Mixed));
512 } else if (BCst && CCst &&
513 ConstantExpr::getAnd(BCst, CCst) == CCst) {
514 result |= (icmp_eq ? FoldMskICmp_BMask_Mixed
515 : FoldMskICmp_BMask_NotMixed);
520 /// Convert an analysis of a masked ICmp into its equivalent if all boolean
521 /// operations had the opposite sense. Since each "NotXXX" flag (recording !=)
522 /// is adjacent to the corresponding normal flag (recording ==), this just
523 /// involves swapping those bits over.
524 static unsigned conjugateICmpMask(unsigned Mask) {
526 NewMask = (Mask & (FoldMskICmp_AMask_AllOnes | FoldMskICmp_BMask_AllOnes |
527 FoldMskICmp_Mask_AllZeroes | FoldMskICmp_AMask_Mixed |
528 FoldMskICmp_BMask_Mixed))
532 (Mask & (FoldMskICmp_AMask_NotAllOnes | FoldMskICmp_BMask_NotAllOnes |
533 FoldMskICmp_Mask_NotAllZeroes | FoldMskICmp_AMask_NotMixed |
534 FoldMskICmp_BMask_NotMixed))
540 /// decomposeBitTestICmp - Decompose an icmp into the form ((X & Y) pred Z)
541 /// if possible. The returned predicate is either == or !=. Returns false if
542 /// decomposition fails.
543 static bool decomposeBitTestICmp(const ICmpInst *I, ICmpInst::Predicate &Pred,
544 Value *&X, Value *&Y, Value *&Z) {
545 ConstantInt *C = dyn_cast<ConstantInt>(I->getOperand(1));
549 switch (I->getPredicate()) {
552 case ICmpInst::ICMP_SLT:
553 // X < 0 is equivalent to (X & SignBit) != 0.
556 Y = ConstantInt::get(I->getContext(), APInt::getSignBit(C->getBitWidth()));
557 Pred = ICmpInst::ICMP_NE;
559 case ICmpInst::ICMP_SGT:
560 // X > -1 is equivalent to (X & SignBit) == 0.
561 if (!C->isAllOnesValue())
563 Y = ConstantInt::get(I->getContext(), APInt::getSignBit(C->getBitWidth()));
564 Pred = ICmpInst::ICMP_EQ;
566 case ICmpInst::ICMP_ULT:
567 // X <u 2^n is equivalent to (X & ~(2^n-1)) == 0.
568 if (!C->getValue().isPowerOf2())
570 Y = ConstantInt::get(I->getContext(), -C->getValue());
571 Pred = ICmpInst::ICMP_EQ;
573 case ICmpInst::ICMP_UGT:
574 // X >u 2^n-1 is equivalent to (X & ~(2^n-1)) != 0.
575 if (!(C->getValue() + 1).isPowerOf2())
577 Y = ConstantInt::get(I->getContext(), ~C->getValue());
578 Pred = ICmpInst::ICMP_NE;
582 X = I->getOperand(0);
583 Z = ConstantInt::getNullValue(C->getType());
587 /// foldLogOpOfMaskedICmpsHelper:
588 /// handle (icmp(A & B) ==/!= C) &/| (icmp(A & D) ==/!= E)
589 /// return the set of pattern classes (from MaskedICmpType)
590 /// that both LHS and RHS satisfy
591 static unsigned foldLogOpOfMaskedICmpsHelper(Value*& A,
592 Value*& B, Value*& C,
593 Value*& D, Value*& E,
594 ICmpInst *LHS, ICmpInst *RHS,
595 ICmpInst::Predicate &LHSCC,
596 ICmpInst::Predicate &RHSCC) {
597 if (LHS->getOperand(0)->getType() != RHS->getOperand(0)->getType()) return 0;
598 // vectors are not (yet?) supported
599 if (LHS->getOperand(0)->getType()->isVectorTy()) return 0;
601 // Here comes the tricky part:
602 // LHS might be of the form L11 & L12 == X, X == L21 & L22,
603 // and L11 & L12 == L21 & L22. The same goes for RHS.
604 // Now we must find those components L** and R**, that are equal, so
605 // that we can extract the parameters A, B, C, D, and E for the canonical
607 Value *L1 = LHS->getOperand(0);
608 Value *L2 = LHS->getOperand(1);
609 Value *L11,*L12,*L21,*L22;
610 // Check whether the icmp can be decomposed into a bit test.
611 if (decomposeBitTestICmp(LHS, LHSCC, L11, L12, L2)) {
612 L21 = L22 = L1 = nullptr;
614 // Look for ANDs in the LHS icmp.
615 if (!L1->getType()->isIntegerTy()) {
616 // You can icmp pointers, for example. They really aren't masks.
618 } else if (!match(L1, m_And(m_Value(L11), m_Value(L12)))) {
619 // Any icmp can be viewed as being trivially masked; if it allows us to
620 // remove one, it's worth it.
622 L12 = Constant::getAllOnesValue(L1->getType());
625 if (!L2->getType()->isIntegerTy()) {
626 // You can icmp pointers, for example. They really aren't masks.
628 } else if (!match(L2, m_And(m_Value(L21), m_Value(L22)))) {
630 L22 = Constant::getAllOnesValue(L2->getType());
634 // Bail if LHS was a icmp that can't be decomposed into an equality.
635 if (!ICmpInst::isEquality(LHSCC))
638 Value *R1 = RHS->getOperand(0);
639 Value *R2 = RHS->getOperand(1);
642 if (decomposeBitTestICmp(RHS, RHSCC, R11, R12, R2)) {
643 if (R11 == L11 || R11 == L12 || R11 == L21 || R11 == L22) {
645 } else if (R12 == L11 || R12 == L12 || R12 == L21 || R12 == L22) {
650 E = R2; R1 = nullptr; ok = true;
651 } else if (R1->getType()->isIntegerTy()) {
652 if (!match(R1, m_And(m_Value(R11), m_Value(R12)))) {
653 // As before, model no mask as a trivial mask if it'll let us do an
656 R12 = Constant::getAllOnesValue(R1->getType());
659 if (R11 == L11 || R11 == L12 || R11 == L21 || R11 == L22) {
660 A = R11; D = R12; E = R2; ok = true;
661 } else if (R12 == L11 || R12 == L12 || R12 == L21 || R12 == L22) {
662 A = R12; D = R11; E = R2; ok = true;
666 // Bail if RHS was a icmp that can't be decomposed into an equality.
667 if (!ICmpInst::isEquality(RHSCC))
670 // Look for ANDs in on the right side of the RHS icmp.
671 if (!ok && R2->getType()->isIntegerTy()) {
672 if (!match(R2, m_And(m_Value(R11), m_Value(R12)))) {
674 R12 = Constant::getAllOnesValue(R2->getType());
677 if (R11 == L11 || R11 == L12 || R11 == L21 || R11 == L22) {
678 A = R11; D = R12; E = R1; ok = true;
679 } else if (R12 == L11 || R12 == L12 || R12 == L21 || R12 == L22) {
680 A = R12; D = R11; E = R1; ok = true;
690 } else if (L12 == A) {
692 } else if (L21 == A) {
694 } else if (L22 == A) {
698 unsigned left_type = getTypeOfMaskedICmp(A, B, C, LHSCC);
699 unsigned right_type = getTypeOfMaskedICmp(A, D, E, RHSCC);
700 return left_type & right_type;
702 /// foldLogOpOfMaskedICmps:
703 /// try to fold (icmp(A & B) ==/!= C) &/| (icmp(A & D) ==/!= E)
704 /// into a single (icmp(A & X) ==/!= Y)
705 static Value *foldLogOpOfMaskedICmps(ICmpInst *LHS, ICmpInst *RHS, bool IsAnd,
706 llvm::InstCombiner::BuilderTy *Builder) {
707 Value *A = nullptr, *B = nullptr, *C = nullptr, *D = nullptr, *E = nullptr;
708 ICmpInst::Predicate LHSCC = LHS->getPredicate(), RHSCC = RHS->getPredicate();
709 unsigned mask = foldLogOpOfMaskedICmpsHelper(A, B, C, D, E, LHS, RHS,
711 if (mask == 0) return nullptr;
712 assert(ICmpInst::isEquality(LHSCC) && ICmpInst::isEquality(RHSCC) &&
713 "foldLogOpOfMaskedICmpsHelper must return an equality predicate.");
715 // In full generality:
716 // (icmp (A & B) Op C) | (icmp (A & D) Op E)
717 // == ![ (icmp (A & B) !Op C) & (icmp (A & D) !Op E) ]
719 // If the latter can be converted into (icmp (A & X) Op Y) then the former is
720 // equivalent to (icmp (A & X) !Op Y).
722 // Therefore, we can pretend for the rest of this function that we're dealing
723 // with the conjunction, provided we flip the sense of any comparisons (both
724 // input and output).
726 // In most cases we're going to produce an EQ for the "&&" case.
727 ICmpInst::Predicate NEWCC = IsAnd ? ICmpInst::ICMP_EQ : ICmpInst::ICMP_NE;
729 // Convert the masking analysis into its equivalent with negated
731 mask = conjugateICmpMask(mask);
734 if (mask & FoldMskICmp_Mask_AllZeroes) {
735 // (icmp eq (A & B), 0) & (icmp eq (A & D), 0)
736 // -> (icmp eq (A & (B|D)), 0)
737 Value *newOr = Builder->CreateOr(B, D);
738 Value *newAnd = Builder->CreateAnd(A, newOr);
739 // we can't use C as zero, because we might actually handle
740 // (icmp ne (A & B), B) & (icmp ne (A & D), D)
741 // with B and D, having a single bit set
742 Value *zero = Constant::getNullValue(A->getType());
743 return Builder->CreateICmp(NEWCC, newAnd, zero);
745 if (mask & FoldMskICmp_BMask_AllOnes) {
746 // (icmp eq (A & B), B) & (icmp eq (A & D), D)
747 // -> (icmp eq (A & (B|D)), (B|D))
748 Value *newOr = Builder->CreateOr(B, D);
749 Value *newAnd = Builder->CreateAnd(A, newOr);
750 return Builder->CreateICmp(NEWCC, newAnd, newOr);
752 if (mask & FoldMskICmp_AMask_AllOnes) {
753 // (icmp eq (A & B), A) & (icmp eq (A & D), A)
754 // -> (icmp eq (A & (B&D)), A)
755 Value *newAnd1 = Builder->CreateAnd(B, D);
756 Value *newAnd = Builder->CreateAnd(A, newAnd1);
757 return Builder->CreateICmp(NEWCC, newAnd, A);
760 // Remaining cases assume at least that B and D are constant, and depend on
761 // their actual values. This isn't strictly, necessary, just a "handle the
762 // easy cases for now" decision.
763 ConstantInt *BCst = dyn_cast<ConstantInt>(B);
764 if (!BCst) return nullptr;
765 ConstantInt *DCst = dyn_cast<ConstantInt>(D);
766 if (!DCst) return nullptr;
768 if (mask & (FoldMskICmp_Mask_NotAllZeroes | FoldMskICmp_BMask_NotAllOnes)) {
769 // (icmp ne (A & B), 0) & (icmp ne (A & D), 0) and
770 // (icmp ne (A & B), B) & (icmp ne (A & D), D)
771 // -> (icmp ne (A & B), 0) or (icmp ne (A & D), 0)
772 // Only valid if one of the masks is a superset of the other (check "B&D" is
773 // the same as either B or D).
774 APInt NewMask = BCst->getValue() & DCst->getValue();
776 if (NewMask == BCst->getValue())
778 else if (NewMask == DCst->getValue())
781 if (mask & FoldMskICmp_AMask_NotAllOnes) {
782 // (icmp ne (A & B), B) & (icmp ne (A & D), D)
783 // -> (icmp ne (A & B), A) or (icmp ne (A & D), A)
784 // Only valid if one of the masks is a superset of the other (check "B|D" is
785 // the same as either B or D).
786 APInt NewMask = BCst->getValue() | DCst->getValue();
788 if (NewMask == BCst->getValue())
790 else if (NewMask == DCst->getValue())
793 if (mask & FoldMskICmp_BMask_Mixed) {
794 // (icmp eq (A & B), C) & (icmp eq (A & D), E)
795 // We already know that B & C == C && D & E == E.
796 // If we can prove that (B & D) & (C ^ E) == 0, that is, the bits of
797 // C and E, which are shared by both the mask B and the mask D, don't
798 // contradict, then we can transform to
799 // -> (icmp eq (A & (B|D)), (C|E))
800 // Currently, we only handle the case of B, C, D, and E being constant.
801 // we can't simply use C and E, because we might actually handle
802 // (icmp ne (A & B), B) & (icmp eq (A & D), D)
803 // with B and D, having a single bit set
804 ConstantInt *CCst = dyn_cast<ConstantInt>(C);
805 if (!CCst) return nullptr;
806 ConstantInt *ECst = dyn_cast<ConstantInt>(E);
807 if (!ECst) return nullptr;
809 CCst = cast<ConstantInt>(ConstantExpr::getXor(BCst, CCst));
811 ECst = cast<ConstantInt>(ConstantExpr::getXor(DCst, ECst));
812 // if there is a conflict we should actually return a false for the
814 if (((BCst->getValue() & DCst->getValue()) &
815 (CCst->getValue() ^ ECst->getValue())) != 0)
816 return ConstantInt::get(LHS->getType(), !IsAnd);
817 Value *newOr1 = Builder->CreateOr(B, D);
818 Value *newOr2 = ConstantExpr::getOr(CCst, ECst);
819 Value *newAnd = Builder->CreateAnd(A, newOr1);
820 return Builder->CreateICmp(NEWCC, newAnd, newOr2);
825 /// Try to fold a signed range checked with lower bound 0 to an unsigned icmp.
826 /// Example: (icmp sge x, 0) & (icmp slt x, n) --> icmp ult x, n
827 /// If \p Inverted is true then the check is for the inverted range, e.g.
828 /// (icmp slt x, 0) | (icmp sgt x, n) --> icmp ugt x, n
829 Value *InstCombiner::simplifyRangeCheck(ICmpInst *Cmp0, ICmpInst *Cmp1,
831 // Check the lower range comparison, e.g. x >= 0
832 // InstCombine already ensured that if there is a constant it's on the RHS.
833 ConstantInt *RangeStart = dyn_cast<ConstantInt>(Cmp0->getOperand(1));
837 ICmpInst::Predicate Pred0 = (Inverted ? Cmp0->getInversePredicate() :
838 Cmp0->getPredicate());
840 // Accept x > -1 or x >= 0 (after potentially inverting the predicate).
841 if (!((Pred0 == ICmpInst::ICMP_SGT && RangeStart->isMinusOne()) ||
842 (Pred0 == ICmpInst::ICMP_SGE && RangeStart->isZero())))
845 ICmpInst::Predicate Pred1 = (Inverted ? Cmp1->getInversePredicate() :
846 Cmp1->getPredicate());
848 Value *Input = Cmp0->getOperand(0);
850 if (Cmp1->getOperand(0) == Input) {
851 // For the upper range compare we have: icmp x, n
852 RangeEnd = Cmp1->getOperand(1);
853 } else if (Cmp1->getOperand(1) == Input) {
854 // For the upper range compare we have: icmp n, x
855 RangeEnd = Cmp1->getOperand(0);
856 Pred1 = ICmpInst::getSwappedPredicate(Pred1);
861 // Check the upper range comparison, e.g. x < n
862 ICmpInst::Predicate NewPred;
864 case ICmpInst::ICMP_SLT: NewPred = ICmpInst::ICMP_ULT; break;
865 case ICmpInst::ICMP_SLE: NewPred = ICmpInst::ICMP_ULE; break;
866 default: return nullptr;
869 // This simplification is only valid if the upper range is not negative.
870 bool IsNegative, IsNotNegative;
871 ComputeSignBit(RangeEnd, IsNotNegative, IsNegative, /*Depth=*/0, Cmp1);
876 NewPred = ICmpInst::getInversePredicate(NewPred);
878 return Builder->CreateICmp(NewPred, Input, RangeEnd);
881 /// FoldAndOfICmps - Fold (icmp)&(icmp) if possible.
882 Value *InstCombiner::FoldAndOfICmps(ICmpInst *LHS, ICmpInst *RHS) {
883 ICmpInst::Predicate LHSCC = LHS->getPredicate(), RHSCC = RHS->getPredicate();
885 // (icmp1 A, B) & (icmp2 A, B) --> (icmp3 A, B)
886 if (PredicatesFoldable(LHSCC, RHSCC)) {
887 if (LHS->getOperand(0) == RHS->getOperand(1) &&
888 LHS->getOperand(1) == RHS->getOperand(0))
890 if (LHS->getOperand(0) == RHS->getOperand(0) &&
891 LHS->getOperand(1) == RHS->getOperand(1)) {
892 Value *Op0 = LHS->getOperand(0), *Op1 = LHS->getOperand(1);
893 unsigned Code = getICmpCode(LHS) & getICmpCode(RHS);
894 bool isSigned = LHS->isSigned() || RHS->isSigned();
895 return getNewICmpValue(isSigned, Code, Op0, Op1, Builder);
899 // handle (roughly): (icmp eq (A & B), C) & (icmp eq (A & D), E)
900 if (Value *V = foldLogOpOfMaskedICmps(LHS, RHS, true, Builder))
903 // E.g. (icmp sge x, 0) & (icmp slt x, n) --> icmp ult x, n
904 if (Value *V = simplifyRangeCheck(LHS, RHS, /*Inverted=*/false))
907 // E.g. (icmp slt x, n) & (icmp sge x, 0) --> icmp ult x, n
908 if (Value *V = simplifyRangeCheck(RHS, LHS, /*Inverted=*/false))
911 // This only handles icmp of constants: (icmp1 A, C1) & (icmp2 B, C2).
912 Value *Val = LHS->getOperand(0), *Val2 = RHS->getOperand(0);
913 ConstantInt *LHSCst = dyn_cast<ConstantInt>(LHS->getOperand(1));
914 ConstantInt *RHSCst = dyn_cast<ConstantInt>(RHS->getOperand(1));
915 if (!LHSCst || !RHSCst) return nullptr;
917 if (LHSCst == RHSCst && LHSCC == RHSCC) {
918 // (icmp ult A, C) & (icmp ult B, C) --> (icmp ult (A|B), C)
919 // where C is a power of 2
920 if (LHSCC == ICmpInst::ICMP_ULT &&
921 LHSCst->getValue().isPowerOf2()) {
922 Value *NewOr = Builder->CreateOr(Val, Val2);
923 return Builder->CreateICmp(LHSCC, NewOr, LHSCst);
926 // (icmp eq A, 0) & (icmp eq B, 0) --> (icmp eq (A|B), 0)
927 if (LHSCC == ICmpInst::ICMP_EQ && LHSCst->isZero()) {
928 Value *NewOr = Builder->CreateOr(Val, Val2);
929 return Builder->CreateICmp(LHSCC, NewOr, LHSCst);
933 // (trunc x) == C1 & (and x, CA) == C2 -> (and x, CA|CMAX) == C1|C2
934 // where CMAX is the all ones value for the truncated type,
935 // iff the lower bits of C2 and CA are zero.
936 if (LHSCC == ICmpInst::ICMP_EQ && LHSCC == RHSCC &&
937 LHS->hasOneUse() && RHS->hasOneUse()) {
939 ConstantInt *AndCst, *SmallCst = nullptr, *BigCst = nullptr;
941 // (trunc x) == C1 & (and x, CA) == C2
942 // (and x, CA) == C2 & (trunc x) == C1
943 if (match(Val2, m_Trunc(m_Value(V))) &&
944 match(Val, m_And(m_Specific(V), m_ConstantInt(AndCst)))) {
947 } else if (match(Val, m_Trunc(m_Value(V))) &&
948 match(Val2, m_And(m_Specific(V), m_ConstantInt(AndCst)))) {
953 if (SmallCst && BigCst) {
954 unsigned BigBitSize = BigCst->getType()->getBitWidth();
955 unsigned SmallBitSize = SmallCst->getType()->getBitWidth();
957 // Check that the low bits are zero.
958 APInt Low = APInt::getLowBitsSet(BigBitSize, SmallBitSize);
959 if ((Low & AndCst->getValue()) == 0 && (Low & BigCst->getValue()) == 0) {
960 Value *NewAnd = Builder->CreateAnd(V, Low | AndCst->getValue());
961 APInt N = SmallCst->getValue().zext(BigBitSize) | BigCst->getValue();
962 Value *NewVal = ConstantInt::get(AndCst->getType()->getContext(), N);
963 return Builder->CreateICmp(LHSCC, NewAnd, NewVal);
968 // From here on, we only handle:
969 // (icmp1 A, C1) & (icmp2 A, C2) --> something simpler.
970 if (Val != Val2) return nullptr;
972 // ICMP_[US][GL]E X, CST is folded to ICMP_[US][GL]T elsewhere.
973 if (LHSCC == ICmpInst::ICMP_UGE || LHSCC == ICmpInst::ICMP_ULE ||
974 RHSCC == ICmpInst::ICMP_UGE || RHSCC == ICmpInst::ICMP_ULE ||
975 LHSCC == ICmpInst::ICMP_SGE || LHSCC == ICmpInst::ICMP_SLE ||
976 RHSCC == ICmpInst::ICMP_SGE || RHSCC == ICmpInst::ICMP_SLE)
979 // Make a constant range that's the intersection of the two icmp ranges.
980 // If the intersection is empty, we know that the result is false.
981 ConstantRange LHSRange =
982 ConstantRange::makeICmpRegion(LHSCC, LHSCst->getValue());
983 ConstantRange RHSRange =
984 ConstantRange::makeICmpRegion(RHSCC, RHSCst->getValue());
986 if (LHSRange.intersectWith(RHSRange).isEmptySet())
987 return ConstantInt::get(CmpInst::makeCmpResultType(LHS->getType()), 0);
989 // We can't fold (ugt x, C) & (sgt x, C2).
990 if (!PredicatesFoldable(LHSCC, RHSCC))
993 // Ensure that the larger constant is on the RHS.
995 if (CmpInst::isSigned(LHSCC) ||
996 (ICmpInst::isEquality(LHSCC) &&
997 CmpInst::isSigned(RHSCC)))
998 ShouldSwap = LHSCst->getValue().sgt(RHSCst->getValue());
1000 ShouldSwap = LHSCst->getValue().ugt(RHSCst->getValue());
1003 std::swap(LHS, RHS);
1004 std::swap(LHSCst, RHSCst);
1005 std::swap(LHSCC, RHSCC);
1008 // At this point, we know we have two icmp instructions
1009 // comparing a value against two constants and and'ing the result
1010 // together. Because of the above check, we know that we only have
1011 // icmp eq, icmp ne, icmp [su]lt, and icmp [SU]gt here. We also know
1012 // (from the icmp folding check above), that the two constants
1013 // are not equal and that the larger constant is on the RHS
1014 assert(LHSCst != RHSCst && "Compares not folded above?");
1017 default: llvm_unreachable("Unknown integer condition code!");
1018 case ICmpInst::ICMP_EQ:
1020 default: llvm_unreachable("Unknown integer condition code!");
1021 case ICmpInst::ICMP_NE: // (X == 13 & X != 15) -> X == 13
1022 case ICmpInst::ICMP_ULT: // (X == 13 & X < 15) -> X == 13
1023 case ICmpInst::ICMP_SLT: // (X == 13 & X < 15) -> X == 13
1026 case ICmpInst::ICMP_NE:
1028 default: llvm_unreachable("Unknown integer condition code!");
1029 case ICmpInst::ICMP_ULT:
1030 if (LHSCst == SubOne(RHSCst)) // (X != 13 & X u< 14) -> X < 13
1031 return Builder->CreateICmpULT(Val, LHSCst);
1032 if (LHSCst->isNullValue()) // (X != 0 & X u< 14) -> X-1 u< 13
1033 return InsertRangeTest(Val, AddOne(LHSCst), RHSCst, false, true);
1034 break; // (X != 13 & X u< 15) -> no change
1035 case ICmpInst::ICMP_SLT:
1036 if (LHSCst == SubOne(RHSCst)) // (X != 13 & X s< 14) -> X < 13
1037 return Builder->CreateICmpSLT(Val, LHSCst);
1038 break; // (X != 13 & X s< 15) -> no change
1039 case ICmpInst::ICMP_EQ: // (X != 13 & X == 15) -> X == 15
1040 case ICmpInst::ICMP_UGT: // (X != 13 & X u> 15) -> X u> 15
1041 case ICmpInst::ICMP_SGT: // (X != 13 & X s> 15) -> X s> 15
1043 case ICmpInst::ICMP_NE:
1044 // Special case to get the ordering right when the values wrap around
1046 if (LHSCst->getValue() == 0 && RHSCst->getValue().isAllOnesValue())
1047 std::swap(LHSCst, RHSCst);
1048 if (LHSCst == SubOne(RHSCst)){// (X != 13 & X != 14) -> X-13 >u 1
1049 Constant *AddCST = ConstantExpr::getNeg(LHSCst);
1050 Value *Add = Builder->CreateAdd(Val, AddCST, Val->getName()+".off");
1051 return Builder->CreateICmpUGT(Add, ConstantInt::get(Add->getType(), 1),
1052 Val->getName()+".cmp");
1054 break; // (X != 13 & X != 15) -> no change
1057 case ICmpInst::ICMP_ULT:
1059 default: llvm_unreachable("Unknown integer condition code!");
1060 case ICmpInst::ICMP_EQ: // (X u< 13 & X == 15) -> false
1061 case ICmpInst::ICMP_UGT: // (X u< 13 & X u> 15) -> false
1062 return ConstantInt::get(CmpInst::makeCmpResultType(LHS->getType()), 0);
1063 case ICmpInst::ICMP_SGT: // (X u< 13 & X s> 15) -> no change
1065 case ICmpInst::ICMP_NE: // (X u< 13 & X != 15) -> X u< 13
1066 case ICmpInst::ICMP_ULT: // (X u< 13 & X u< 15) -> X u< 13
1068 case ICmpInst::ICMP_SLT: // (X u< 13 & X s< 15) -> no change
1072 case ICmpInst::ICMP_SLT:
1074 default: llvm_unreachable("Unknown integer condition code!");
1075 case ICmpInst::ICMP_UGT: // (X s< 13 & X u> 15) -> no change
1077 case ICmpInst::ICMP_NE: // (X s< 13 & X != 15) -> X < 13
1078 case ICmpInst::ICMP_SLT: // (X s< 13 & X s< 15) -> X < 13
1080 case ICmpInst::ICMP_ULT: // (X s< 13 & X u< 15) -> no change
1084 case ICmpInst::ICMP_UGT:
1086 default: llvm_unreachable("Unknown integer condition code!");
1087 case ICmpInst::ICMP_EQ: // (X u> 13 & X == 15) -> X == 15
1088 case ICmpInst::ICMP_UGT: // (X u> 13 & X u> 15) -> X u> 15
1090 case ICmpInst::ICMP_SGT: // (X u> 13 & X s> 15) -> no change
1092 case ICmpInst::ICMP_NE:
1093 if (RHSCst == AddOne(LHSCst)) // (X u> 13 & X != 14) -> X u> 14
1094 return Builder->CreateICmp(LHSCC, Val, RHSCst);
1095 break; // (X u> 13 & X != 15) -> no change
1096 case ICmpInst::ICMP_ULT: // (X u> 13 & X u< 15) -> (X-14) <u 1
1097 return InsertRangeTest(Val, AddOne(LHSCst), RHSCst, false, true);
1098 case ICmpInst::ICMP_SLT: // (X u> 13 & X s< 15) -> no change
1102 case ICmpInst::ICMP_SGT:
1104 default: llvm_unreachable("Unknown integer condition code!");
1105 case ICmpInst::ICMP_EQ: // (X s> 13 & X == 15) -> X == 15
1106 case ICmpInst::ICMP_SGT: // (X s> 13 & X s> 15) -> X s> 15
1108 case ICmpInst::ICMP_UGT: // (X s> 13 & X u> 15) -> no change
1110 case ICmpInst::ICMP_NE:
1111 if (RHSCst == AddOne(LHSCst)) // (X s> 13 & X != 14) -> X s> 14
1112 return Builder->CreateICmp(LHSCC, Val, RHSCst);
1113 break; // (X s> 13 & X != 15) -> no change
1114 case ICmpInst::ICMP_SLT: // (X s> 13 & X s< 15) -> (X-14) s< 1
1115 return InsertRangeTest(Val, AddOne(LHSCst), RHSCst, true, true);
1116 case ICmpInst::ICMP_ULT: // (X s> 13 & X u< 15) -> no change
1125 /// FoldAndOfFCmps - Optimize (fcmp)&(fcmp). NOTE: Unlike the rest of
1126 /// instcombine, this returns a Value which should already be inserted into the
1128 Value *InstCombiner::FoldAndOfFCmps(FCmpInst *LHS, FCmpInst *RHS) {
1129 if (LHS->getPredicate() == FCmpInst::FCMP_ORD &&
1130 RHS->getPredicate() == FCmpInst::FCMP_ORD) {
1131 if (LHS->getOperand(0)->getType() != RHS->getOperand(0)->getType())
1134 // (fcmp ord x, c) & (fcmp ord y, c) -> (fcmp ord x, y)
1135 if (ConstantFP *LHSC = dyn_cast<ConstantFP>(LHS->getOperand(1)))
1136 if (ConstantFP *RHSC = dyn_cast<ConstantFP>(RHS->getOperand(1))) {
1137 // If either of the constants are nans, then the whole thing returns
1139 if (LHSC->getValueAPF().isNaN() || RHSC->getValueAPF().isNaN())
1140 return Builder->getFalse();
1141 return Builder->CreateFCmpORD(LHS->getOperand(0), RHS->getOperand(0));
1144 // Handle vector zeros. This occurs because the canonical form of
1145 // "fcmp ord x,x" is "fcmp ord x, 0".
1146 if (isa<ConstantAggregateZero>(LHS->getOperand(1)) &&
1147 isa<ConstantAggregateZero>(RHS->getOperand(1)))
1148 return Builder->CreateFCmpORD(LHS->getOperand(0), RHS->getOperand(0));
1152 Value *Op0LHS = LHS->getOperand(0), *Op0RHS = LHS->getOperand(1);
1153 Value *Op1LHS = RHS->getOperand(0), *Op1RHS = RHS->getOperand(1);
1154 FCmpInst::Predicate Op0CC = LHS->getPredicate(), Op1CC = RHS->getPredicate();
1157 if (Op0LHS == Op1RHS && Op0RHS == Op1LHS) {
1158 // Swap RHS operands to match LHS.
1159 Op1CC = FCmpInst::getSwappedPredicate(Op1CC);
1160 std::swap(Op1LHS, Op1RHS);
1163 if (Op0LHS == Op1LHS && Op0RHS == Op1RHS) {
1164 // Simplify (fcmp cc0 x, y) & (fcmp cc1 x, y).
1166 return Builder->CreateFCmp((FCmpInst::Predicate)Op0CC, Op0LHS, Op0RHS);
1167 if (Op0CC == FCmpInst::FCMP_FALSE || Op1CC == FCmpInst::FCMP_FALSE)
1168 return ConstantInt::get(CmpInst::makeCmpResultType(LHS->getType()), 0);
1169 if (Op0CC == FCmpInst::FCMP_TRUE)
1171 if (Op1CC == FCmpInst::FCMP_TRUE)
1176 unsigned Op0Pred = getFCmpCode(Op0CC, Op0Ordered);
1177 unsigned Op1Pred = getFCmpCode(Op1CC, Op1Ordered);
1178 // uno && ord -> false
1179 if (Op0Pred == 0 && Op1Pred == 0 && Op0Ordered != Op1Ordered)
1180 return ConstantInt::get(CmpInst::makeCmpResultType(LHS->getType()), 0);
1182 std::swap(LHS, RHS);
1183 std::swap(Op0Pred, Op1Pred);
1184 std::swap(Op0Ordered, Op1Ordered);
1187 // uno && ueq -> uno && (uno || eq) -> uno
1188 // ord && olt -> ord && (ord && lt) -> olt
1189 if (!Op0Ordered && (Op0Ordered == Op1Ordered))
1191 if (Op0Ordered && (Op0Ordered == Op1Ordered))
1194 // uno && oeq -> uno && (ord && eq) -> false
1196 return ConstantInt::get(CmpInst::makeCmpResultType(LHS->getType()), 0);
1197 // ord && ueq -> ord && (uno || eq) -> oeq
1198 return getFCmpValue(true, Op1Pred, Op0LHS, Op0RHS, Builder);
1205 Instruction *InstCombiner::visitAnd(BinaryOperator &I) {
1206 bool Changed = SimplifyAssociativeOrCommutative(I);
1207 Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
1209 if (Value *V = SimplifyVectorOp(I))
1210 return ReplaceInstUsesWith(I, V);
1212 if (Value *V = SimplifyAndInst(Op0, Op1, DL, TLI, DT, AC))
1213 return ReplaceInstUsesWith(I, V);
1215 // (A|B)&(A|C) -> A|(B&C) etc
1216 if (Value *V = SimplifyUsingDistributiveLaws(I))
1217 return ReplaceInstUsesWith(I, V);
1219 // See if we can simplify any instructions used by the instruction whose sole
1220 // purpose is to compute bits we don't care about.
1221 if (SimplifyDemandedInstructionBits(I))
1224 if (Value *V = SimplifyBSwap(I))
1225 return ReplaceInstUsesWith(I, V);
1227 if (ConstantInt *AndRHS = dyn_cast<ConstantInt>(Op1)) {
1228 const APInt &AndRHSMask = AndRHS->getValue();
1230 // Optimize a variety of ((val OP C1) & C2) combinations...
1231 if (BinaryOperator *Op0I = dyn_cast<BinaryOperator>(Op0)) {
1232 Value *Op0LHS = Op0I->getOperand(0);
1233 Value *Op0RHS = Op0I->getOperand(1);
1234 switch (Op0I->getOpcode()) {
1236 case Instruction::Xor:
1237 case Instruction::Or: {
1238 // If the mask is only needed on one incoming arm, push it up.
1239 if (!Op0I->hasOneUse()) break;
1241 APInt NotAndRHS(~AndRHSMask);
1242 if (MaskedValueIsZero(Op0LHS, NotAndRHS, 0, &I)) {
1243 // Not masking anything out for the LHS, move to RHS.
1244 Value *NewRHS = Builder->CreateAnd(Op0RHS, AndRHS,
1245 Op0RHS->getName()+".masked");
1246 return BinaryOperator::Create(Op0I->getOpcode(), Op0LHS, NewRHS);
1248 if (!isa<Constant>(Op0RHS) &&
1249 MaskedValueIsZero(Op0RHS, NotAndRHS, 0, &I)) {
1250 // Not masking anything out for the RHS, move to LHS.
1251 Value *NewLHS = Builder->CreateAnd(Op0LHS, AndRHS,
1252 Op0LHS->getName()+".masked");
1253 return BinaryOperator::Create(Op0I->getOpcode(), NewLHS, Op0RHS);
1258 case Instruction::Add:
1259 // ((A & N) + B) & AndRHS -> (A + B) & AndRHS iff N&AndRHS == AndRHS.
1260 // ((A | N) + B) & AndRHS -> (A + B) & AndRHS iff N&AndRHS == 0
1261 // ((A ^ N) + B) & AndRHS -> (A + B) & AndRHS iff N&AndRHS == 0
1262 if (Value *V = FoldLogicalPlusAnd(Op0LHS, Op0RHS, AndRHS, false, I))
1263 return BinaryOperator::CreateAnd(V, AndRHS);
1264 if (Value *V = FoldLogicalPlusAnd(Op0RHS, Op0LHS, AndRHS, false, I))
1265 return BinaryOperator::CreateAnd(V, AndRHS); // Add commutes
1268 case Instruction::Sub:
1269 // ((A & N) - B) & AndRHS -> (A - B) & AndRHS iff N&AndRHS == AndRHS.
1270 // ((A | N) - B) & AndRHS -> (A - B) & AndRHS iff N&AndRHS == 0
1271 // ((A ^ N) - B) & AndRHS -> (A - B) & AndRHS iff N&AndRHS == 0
1272 if (Value *V = FoldLogicalPlusAnd(Op0LHS, Op0RHS, AndRHS, true, I))
1273 return BinaryOperator::CreateAnd(V, AndRHS);
1275 // (A - N) & AndRHS -> -N & AndRHS iff A&AndRHS==0 and AndRHS
1276 // has 1's for all bits that the subtraction with A might affect.
1277 if (Op0I->hasOneUse() && !match(Op0LHS, m_Zero())) {
1278 uint32_t BitWidth = AndRHSMask.getBitWidth();
1279 uint32_t Zeros = AndRHSMask.countLeadingZeros();
1280 APInt Mask = APInt::getLowBitsSet(BitWidth, BitWidth - Zeros);
1282 if (MaskedValueIsZero(Op0LHS, Mask, 0, &I)) {
1283 Value *NewNeg = Builder->CreateNeg(Op0RHS);
1284 return BinaryOperator::CreateAnd(NewNeg, AndRHS);
1289 case Instruction::Shl:
1290 case Instruction::LShr:
1291 // (1 << x) & 1 --> zext(x == 0)
1292 // (1 >> x) & 1 --> zext(x == 0)
1293 if (AndRHSMask == 1 && Op0LHS == AndRHS) {
1295 Builder->CreateICmpEQ(Op0RHS, Constant::getNullValue(I.getType()));
1296 return new ZExtInst(NewICmp, I.getType());
1301 if (ConstantInt *Op0CI = dyn_cast<ConstantInt>(Op0I->getOperand(1)))
1302 if (Instruction *Res = OptAndOp(Op0I, Op0CI, AndRHS, I))
1306 // If this is an integer truncation, and if the source is an 'and' with
1307 // immediate, transform it. This frequently occurs for bitfield accesses.
1309 Value *X = nullptr; ConstantInt *YC = nullptr;
1310 if (match(Op0, m_Trunc(m_And(m_Value(X), m_ConstantInt(YC))))) {
1311 // Change: and (trunc (and X, YC) to T), C2
1312 // into : and (trunc X to T), trunc(YC) & C2
1313 // This will fold the two constants together, which may allow
1314 // other simplifications.
1315 Value *NewCast = Builder->CreateTrunc(X, I.getType(), "and.shrunk");
1316 Constant *C3 = ConstantExpr::getTrunc(YC, I.getType());
1317 C3 = ConstantExpr::getAnd(C3, AndRHS);
1318 return BinaryOperator::CreateAnd(NewCast, C3);
1322 // Try to fold constant and into select arguments.
1323 if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
1324 if (Instruction *R = FoldOpIntoSelect(I, SI))
1326 if (isa<PHINode>(Op0))
1327 if (Instruction *NV = FoldOpIntoPhi(I))
1332 // (~A & ~B) == (~(A | B)) - De Morgan's Law
1333 if (Value *Op0NotVal = dyn_castNotVal(Op0))
1334 if (Value *Op1NotVal = dyn_castNotVal(Op1))
1335 if (Op0->hasOneUse() && Op1->hasOneUse()) {
1336 Value *Or = Builder->CreateOr(Op0NotVal, Op1NotVal,
1337 I.getName()+".demorgan");
1338 return BinaryOperator::CreateNot(Or);
1342 Value *A = nullptr, *B = nullptr, *C = nullptr, *D = nullptr;
1343 // (A|B) & ~(A&B) -> A^B
1344 if (match(Op0, m_Or(m_Value(A), m_Value(B))) &&
1345 match(Op1, m_Not(m_And(m_Value(C), m_Value(D)))) &&
1346 ((A == C && B == D) || (A == D && B == C)))
1347 return BinaryOperator::CreateXor(A, B);
1349 // ~(A&B) & (A|B) -> A^B
1350 if (match(Op1, m_Or(m_Value(A), m_Value(B))) &&
1351 match(Op0, m_Not(m_And(m_Value(C), m_Value(D)))) &&
1352 ((A == C && B == D) || (A == D && B == C)))
1353 return BinaryOperator::CreateXor(A, B);
1355 // A&(A^B) => A & ~B
1357 Value *tmpOp0 = Op0;
1358 Value *tmpOp1 = Op1;
1359 if (Op0->hasOneUse() &&
1360 match(Op0, m_Xor(m_Value(A), m_Value(B)))) {
1361 if (A == Op1 || B == Op1 ) {
1368 if (tmpOp1->hasOneUse() &&
1369 match(tmpOp1, m_Xor(m_Value(A), m_Value(B)))) {
1373 // Notice that the patten (A&(~B)) is actually (A&(-1^B)), so if
1374 // A is originally -1 (or a vector of -1 and undefs), then we enter
1375 // an endless loop. By checking that A is non-constant we ensure that
1376 // we will never get to the loop.
1377 if (A == tmpOp0 && !isa<Constant>(A)) // A&(A^B) -> A & ~B
1378 return BinaryOperator::CreateAnd(A, Builder->CreateNot(B));
1382 // (A&((~A)|B)) -> A&B
1383 if (match(Op0, m_Or(m_Not(m_Specific(Op1)), m_Value(A))) ||
1384 match(Op0, m_Or(m_Value(A), m_Not(m_Specific(Op1)))))
1385 return BinaryOperator::CreateAnd(A, Op1);
1386 if (match(Op1, m_Or(m_Not(m_Specific(Op0)), m_Value(A))) ||
1387 match(Op1, m_Or(m_Value(A), m_Not(m_Specific(Op0)))))
1388 return BinaryOperator::CreateAnd(A, Op0);
1390 // (A ^ B) & ((B ^ C) ^ A) -> (A ^ B) & ~C
1391 if (match(Op0, m_Xor(m_Value(A), m_Value(B))))
1392 if (match(Op1, m_Xor(m_Xor(m_Specific(B), m_Value(C)), m_Specific(A))))
1393 if (Op1->hasOneUse() || cast<BinaryOperator>(Op1)->hasOneUse())
1394 return BinaryOperator::CreateAnd(Op0, Builder->CreateNot(C));
1396 // ((A ^ C) ^ B) & (B ^ A) -> (B ^ A) & ~C
1397 if (match(Op0, m_Xor(m_Xor(m_Value(A), m_Value(C)), m_Value(B))))
1398 if (match(Op1, m_Xor(m_Specific(B), m_Specific(A))))
1399 if (Op0->hasOneUse() || cast<BinaryOperator>(Op0)->hasOneUse())
1400 return BinaryOperator::CreateAnd(Op1, Builder->CreateNot(C));
1402 // (A | B) & ((~A) ^ B) -> (A & B)
1403 if (match(Op0, m_Or(m_Value(A), m_Value(B))) &&
1404 match(Op1, m_Xor(m_Not(m_Specific(A)), m_Specific(B))))
1405 return BinaryOperator::CreateAnd(A, B);
1407 // ((~A) ^ B) & (A | B) -> (A & B)
1408 if (match(Op0, m_Xor(m_Not(m_Value(A)), m_Value(B))) &&
1409 match(Op1, m_Or(m_Specific(A), m_Specific(B))))
1410 return BinaryOperator::CreateAnd(A, B);
1414 ICmpInst *LHS = dyn_cast<ICmpInst>(Op0);
1415 ICmpInst *RHS = dyn_cast<ICmpInst>(Op1);
1417 if (Value *Res = FoldAndOfICmps(LHS, RHS))
1418 return ReplaceInstUsesWith(I, Res);
1420 // TODO: Make this recursive; it's a little tricky because an arbitrary
1421 // number of 'and' instructions might have to be created.
1423 if (LHS && match(Op1, m_OneUse(m_And(m_Value(X), m_Value(Y))))) {
1424 if (auto *Cmp = dyn_cast<ICmpInst>(X))
1425 if (Value *Res = FoldAndOfICmps(LHS, Cmp))
1426 return ReplaceInstUsesWith(I, Builder->CreateAnd(Res, Y));
1427 if (auto *Cmp = dyn_cast<ICmpInst>(Y))
1428 if (Value *Res = FoldAndOfICmps(LHS, Cmp))
1429 return ReplaceInstUsesWith(I, Builder->CreateAnd(Res, X));
1431 if (RHS && match(Op0, m_OneUse(m_And(m_Value(X), m_Value(Y))))) {
1432 if (auto *Cmp = dyn_cast<ICmpInst>(X))
1433 if (Value *Res = FoldAndOfICmps(Cmp, RHS))
1434 return ReplaceInstUsesWith(I, Builder->CreateAnd(Res, Y));
1435 if (auto *Cmp = dyn_cast<ICmpInst>(Y))
1436 if (Value *Res = FoldAndOfICmps(Cmp, RHS))
1437 return ReplaceInstUsesWith(I, Builder->CreateAnd(Res, X));
1441 // If and'ing two fcmp, try combine them into one.
1442 if (FCmpInst *LHS = dyn_cast<FCmpInst>(I.getOperand(0)))
1443 if (FCmpInst *RHS = dyn_cast<FCmpInst>(I.getOperand(1)))
1444 if (Value *Res = FoldAndOfFCmps(LHS, RHS))
1445 return ReplaceInstUsesWith(I, Res);
1448 // fold (and (cast A), (cast B)) -> (cast (and A, B))
1449 if (CastInst *Op0C = dyn_cast<CastInst>(Op0))
1450 if (CastInst *Op1C = dyn_cast<CastInst>(Op1)) {
1451 Type *SrcTy = Op0C->getOperand(0)->getType();
1452 if (Op0C->getOpcode() == Op1C->getOpcode() && // same cast kind ?
1453 SrcTy == Op1C->getOperand(0)->getType() &&
1454 SrcTy->isIntOrIntVectorTy()) {
1455 Value *Op0COp = Op0C->getOperand(0), *Op1COp = Op1C->getOperand(0);
1457 // Only do this if the casts both really cause code to be generated.
1458 if (ShouldOptimizeCast(Op0C->getOpcode(), Op0COp, I.getType()) &&
1459 ShouldOptimizeCast(Op1C->getOpcode(), Op1COp, I.getType())) {
1460 Value *NewOp = Builder->CreateAnd(Op0COp, Op1COp, I.getName());
1461 return CastInst::Create(Op0C->getOpcode(), NewOp, I.getType());
1464 // If this is and(cast(icmp), cast(icmp)), try to fold this even if the
1465 // cast is otherwise not optimizable. This happens for vector sexts.
1466 if (ICmpInst *RHS = dyn_cast<ICmpInst>(Op1COp))
1467 if (ICmpInst *LHS = dyn_cast<ICmpInst>(Op0COp))
1468 if (Value *Res = FoldAndOfICmps(LHS, RHS))
1469 return CastInst::Create(Op0C->getOpcode(), Res, I.getType());
1471 // If this is and(cast(fcmp), cast(fcmp)), try to fold this even if the
1472 // cast is otherwise not optimizable. This happens for vector sexts.
1473 if (FCmpInst *RHS = dyn_cast<FCmpInst>(Op1COp))
1474 if (FCmpInst *LHS = dyn_cast<FCmpInst>(Op0COp))
1475 if (Value *Res = FoldAndOfFCmps(LHS, RHS))
1476 return CastInst::Create(Op0C->getOpcode(), Res, I.getType());
1482 bool OpsSwapped = false;
1483 // Canonicalize SExt or Not to the LHS
1484 if (match(Op1, m_SExt(m_Value())) ||
1485 match(Op1, m_Not(m_Value()))) {
1486 std::swap(Op0, Op1);
1490 // Fold (and (sext bool to A), B) --> (select bool, B, 0)
1491 if (match(Op0, m_SExt(m_Value(X))) &&
1492 X->getType()->getScalarType()->isIntegerTy(1)) {
1493 Value *Zero = Constant::getNullValue(Op1->getType());
1494 return SelectInst::Create(X, Op1, Zero);
1497 // Fold (and ~(sext bool to A), B) --> (select bool, 0, B)
1498 if (match(Op0, m_Not(m_SExt(m_Value(X)))) &&
1499 X->getType()->getScalarType()->isIntegerTy(1)) {
1500 Value *Zero = Constant::getNullValue(Op0->getType());
1501 return SelectInst::Create(X, Zero, Op1);
1505 std::swap(Op0, Op1);
1508 return Changed ? &I : nullptr;
1511 /// CollectBSwapParts - Analyze the specified subexpression and see if it is
1512 /// capable of providing pieces of a bswap. The subexpression provides pieces
1513 /// of a bswap if it is proven that each of the non-zero bytes in the output of
1514 /// the expression came from the corresponding "byte swapped" byte in some other
1515 /// value. For example, if the current subexpression is "(shl i32 %X, 24)" then
1516 /// we know that the expression deposits the low byte of %X into the high byte
1517 /// of the bswap result and that all other bytes are zero. This expression is
1518 /// accepted, the high byte of ByteValues is set to X to indicate a correct
1521 /// This function returns true if the match was unsuccessful and false if so.
1522 /// On entry to the function the "OverallLeftShift" is a signed integer value
1523 /// indicating the number of bytes that the subexpression is later shifted. For
1524 /// example, if the expression is later right shifted by 16 bits, the
1525 /// OverallLeftShift value would be -2 on entry. This is used to specify which
1526 /// byte of ByteValues is actually being set.
1528 /// Similarly, ByteMask is a bitmask where a bit is clear if its corresponding
1529 /// byte is masked to zero by a user. For example, in (X & 255), X will be
1530 /// processed with a bytemask of 1. Because bytemask is 32-bits, this limits
1531 /// this function to working on up to 32-byte (256 bit) values. ByteMask is
1532 /// always in the local (OverallLeftShift) coordinate space.
1534 static bool CollectBSwapParts(Value *V, int OverallLeftShift, uint32_t ByteMask,
1535 SmallVectorImpl<Value *> &ByteValues) {
1536 if (Instruction *I = dyn_cast<Instruction>(V)) {
1537 // If this is an or instruction, it may be an inner node of the bswap.
1538 if (I->getOpcode() == Instruction::Or) {
1539 return CollectBSwapParts(I->getOperand(0), OverallLeftShift, ByteMask,
1541 CollectBSwapParts(I->getOperand(1), OverallLeftShift, ByteMask,
1545 // If this is a logical shift by a constant multiple of 8, recurse with
1546 // OverallLeftShift and ByteMask adjusted.
1547 if (I->isLogicalShift() && isa<ConstantInt>(I->getOperand(1))) {
1549 cast<ConstantInt>(I->getOperand(1))->getLimitedValue(~0U);
1550 // Ensure the shift amount is defined and of a byte value.
1551 if ((ShAmt & 7) || (ShAmt > 8*ByteValues.size()))
1554 unsigned ByteShift = ShAmt >> 3;
1555 if (I->getOpcode() == Instruction::Shl) {
1556 // X << 2 -> collect(X, +2)
1557 OverallLeftShift += ByteShift;
1558 ByteMask >>= ByteShift;
1560 // X >>u 2 -> collect(X, -2)
1561 OverallLeftShift -= ByteShift;
1562 ByteMask <<= ByteShift;
1563 ByteMask &= (~0U >> (32-ByteValues.size()));
1566 if (OverallLeftShift >= (int)ByteValues.size()) return true;
1567 if (OverallLeftShift <= -(int)ByteValues.size()) return true;
1569 return CollectBSwapParts(I->getOperand(0), OverallLeftShift, ByteMask,
1573 // If this is a logical 'and' with a mask that clears bytes, clear the
1574 // corresponding bytes in ByteMask.
1575 if (I->getOpcode() == Instruction::And &&
1576 isa<ConstantInt>(I->getOperand(1))) {
1577 // Scan every byte of the and mask, seeing if the byte is either 0 or 255.
1578 unsigned NumBytes = ByteValues.size();
1579 APInt Byte(I->getType()->getPrimitiveSizeInBits(), 255);
1580 const APInt &AndMask = cast<ConstantInt>(I->getOperand(1))->getValue();
1582 for (unsigned i = 0; i != NumBytes; ++i, Byte <<= 8) {
1583 // If this byte is masked out by a later operation, we don't care what
1585 if ((ByteMask & (1 << i)) == 0)
1588 // If the AndMask is all zeros for this byte, clear the bit.
1589 APInt MaskB = AndMask & Byte;
1591 ByteMask &= ~(1U << i);
1595 // If the AndMask is not all ones for this byte, it's not a bytezap.
1599 // Otherwise, this byte is kept.
1602 return CollectBSwapParts(I->getOperand(0), OverallLeftShift, ByteMask,
1607 // Okay, we got to something that isn't a shift, 'or' or 'and'. This must be
1608 // the input value to the bswap. Some observations: 1) if more than one byte
1609 // is demanded from this input, then it could not be successfully assembled
1610 // into a byteswap. At least one of the two bytes would not be aligned with
1611 // their ultimate destination.
1612 if (!isPowerOf2_32(ByteMask)) return true;
1613 unsigned InputByteNo = countTrailingZeros(ByteMask);
1615 // 2) The input and ultimate destinations must line up: if byte 3 of an i32
1616 // is demanded, it needs to go into byte 0 of the result. This means that the
1617 // byte needs to be shifted until it lands in the right byte bucket. The
1618 // shift amount depends on the position: if the byte is coming from the high
1619 // part of the value (e.g. byte 3) then it must be shifted right. If from the
1620 // low part, it must be shifted left.
1621 unsigned DestByteNo = InputByteNo + OverallLeftShift;
1622 if (ByteValues.size()-1-DestByteNo != InputByteNo)
1625 // If the destination byte value is already defined, the values are or'd
1626 // together, which isn't a bswap (unless it's an or of the same bits).
1627 if (ByteValues[DestByteNo] && ByteValues[DestByteNo] != V)
1629 ByteValues[DestByteNo] = V;
1633 /// MatchBSwap - Given an OR instruction, check to see if this is a bswap idiom.
1634 /// If so, insert the new bswap intrinsic and return it.
1635 Instruction *InstCombiner::MatchBSwap(BinaryOperator &I) {
1636 IntegerType *ITy = dyn_cast<IntegerType>(I.getType());
1637 if (!ITy || ITy->getBitWidth() % 16 ||
1638 // ByteMask only allows up to 32-byte values.
1639 ITy->getBitWidth() > 32*8)
1640 return nullptr; // Can only bswap pairs of bytes. Can't do vectors.
1642 /// ByteValues - For each byte of the result, we keep track of which value
1643 /// defines each byte.
1644 SmallVector<Value*, 8> ByteValues;
1645 ByteValues.resize(ITy->getBitWidth()/8);
1647 // Try to find all the pieces corresponding to the bswap.
1648 uint32_t ByteMask = ~0U >> (32-ByteValues.size());
1649 if (CollectBSwapParts(&I, 0, ByteMask, ByteValues))
1652 // Check to see if all of the bytes come from the same value.
1653 Value *V = ByteValues[0];
1654 if (!V) return nullptr; // Didn't find a byte? Must be zero.
1656 // Check to make sure that all of the bytes come from the same value.
1657 for (unsigned i = 1, e = ByteValues.size(); i != e; ++i)
1658 if (ByteValues[i] != V)
1660 Module *M = I.getParent()->getParent()->getParent();
1661 Function *F = Intrinsic::getDeclaration(M, Intrinsic::bswap, ITy);
1662 return CallInst::Create(F, V);
1665 /// MatchSelectFromAndOr - We have an expression of the form (A&C)|(B&D). Check
1666 /// If A is (cond?-1:0) and either B or D is ~(cond?-1,0) or (cond?0,-1), then
1667 /// we can simplify this expression to "cond ? C : D or B".
1668 static Instruction *MatchSelectFromAndOr(Value *A, Value *B,
1669 Value *C, Value *D) {
1670 // If A is not a select of -1/0, this cannot match.
1671 Value *Cond = nullptr;
1672 if (!match(A, m_SExt(m_Value(Cond))) ||
1673 !Cond->getType()->isIntegerTy(1))
1676 // ((cond?-1:0)&C) | (B&(cond?0:-1)) -> cond ? C : B.
1677 if (match(D, m_Not(m_SExt(m_Specific(Cond)))))
1678 return SelectInst::Create(Cond, C, B);
1679 if (match(D, m_SExt(m_Not(m_Specific(Cond)))))
1680 return SelectInst::Create(Cond, C, B);
1682 // ((cond?-1:0)&C) | ((cond?0:-1)&D) -> cond ? C : D.
1683 if (match(B, m_Not(m_SExt(m_Specific(Cond)))))
1684 return SelectInst::Create(Cond, C, D);
1685 if (match(B, m_SExt(m_Not(m_Specific(Cond)))))
1686 return SelectInst::Create(Cond, C, D);
1690 /// FoldOrOfICmps - Fold (icmp)|(icmp) if possible.
1691 Value *InstCombiner::FoldOrOfICmps(ICmpInst *LHS, ICmpInst *RHS,
1692 Instruction *CxtI) {
1693 ICmpInst::Predicate LHSCC = LHS->getPredicate(), RHSCC = RHS->getPredicate();
1695 // Fold (iszero(A & K1) | iszero(A & K2)) -> (A & (K1 | K2)) != (K1 | K2)
1696 // if K1 and K2 are a one-bit mask.
1697 ConstantInt *LHSCst = dyn_cast<ConstantInt>(LHS->getOperand(1));
1698 ConstantInt *RHSCst = dyn_cast<ConstantInt>(RHS->getOperand(1));
1700 if (LHS->getPredicate() == ICmpInst::ICMP_EQ && LHSCst && LHSCst->isZero() &&
1701 RHS->getPredicate() == ICmpInst::ICMP_EQ && RHSCst && RHSCst->isZero()) {
1703 BinaryOperator *LAnd = dyn_cast<BinaryOperator>(LHS->getOperand(0));
1704 BinaryOperator *RAnd = dyn_cast<BinaryOperator>(RHS->getOperand(0));
1705 if (LAnd && RAnd && LAnd->hasOneUse() && RHS->hasOneUse() &&
1706 LAnd->getOpcode() == Instruction::And &&
1707 RAnd->getOpcode() == Instruction::And) {
1709 Value *Mask = nullptr;
1710 Value *Masked = nullptr;
1711 if (LAnd->getOperand(0) == RAnd->getOperand(0) &&
1712 isKnownToBeAPowerOfTwo(LAnd->getOperand(1), false, 0, AC, CxtI, DT) &&
1713 isKnownToBeAPowerOfTwo(RAnd->getOperand(1), false, 0, AC, CxtI, DT)) {
1714 Mask = Builder->CreateOr(LAnd->getOperand(1), RAnd->getOperand(1));
1715 Masked = Builder->CreateAnd(LAnd->getOperand(0), Mask);
1716 } else if (LAnd->getOperand(1) == RAnd->getOperand(1) &&
1717 isKnownToBeAPowerOfTwo(LAnd->getOperand(0), false, 0, AC, CxtI,
1719 isKnownToBeAPowerOfTwo(RAnd->getOperand(0), false, 0, AC, CxtI,
1721 Mask = Builder->CreateOr(LAnd->getOperand(0), RAnd->getOperand(0));
1722 Masked = Builder->CreateAnd(LAnd->getOperand(1), Mask);
1726 return Builder->CreateICmp(ICmpInst::ICMP_NE, Masked, Mask);
1730 // Fold (icmp ult/ule (A + C1), C3) | (icmp ult/ule (A + C2), C3)
1731 // --> (icmp ult/ule ((A & ~(C1 ^ C2)) + max(C1, C2)), C3)
1732 // The original condition actually refers to the following two ranges:
1733 // [MAX_UINT-C1+1, MAX_UINT-C1+1+C3] and [MAX_UINT-C2+1, MAX_UINT-C2+1+C3]
1734 // We can fold these two ranges if:
1735 // 1) C1 and C2 is unsigned greater than C3.
1736 // 2) The two ranges are separated.
1737 // 3) C1 ^ C2 is one-bit mask.
1738 // 4) LowRange1 ^ LowRange2 and HighRange1 ^ HighRange2 are one-bit mask.
1739 // This implies all values in the two ranges differ by exactly one bit.
1741 if ((LHSCC == ICmpInst::ICMP_ULT || LHSCC == ICmpInst::ICMP_ULE) &&
1742 LHSCC == RHSCC && LHSCst && RHSCst && LHS->hasOneUse() &&
1743 RHS->hasOneUse() && LHSCst->getType() == RHSCst->getType() &&
1744 LHSCst->getValue() == (RHSCst->getValue())) {
1746 Value *LAdd = LHS->getOperand(0);
1747 Value *RAdd = RHS->getOperand(0);
1749 Value *LAddOpnd, *RAddOpnd;
1750 ConstantInt *LAddCst, *RAddCst;
1751 if (match(LAdd, m_Add(m_Value(LAddOpnd), m_ConstantInt(LAddCst))) &&
1752 match(RAdd, m_Add(m_Value(RAddOpnd), m_ConstantInt(RAddCst))) &&
1753 LAddCst->getValue().ugt(LHSCst->getValue()) &&
1754 RAddCst->getValue().ugt(LHSCst->getValue())) {
1756 APInt DiffCst = LAddCst->getValue() ^ RAddCst->getValue();
1757 if (LAddOpnd == RAddOpnd && DiffCst.isPowerOf2()) {
1758 ConstantInt *MaxAddCst = nullptr;
1759 if (LAddCst->getValue().ult(RAddCst->getValue()))
1760 MaxAddCst = RAddCst;
1762 MaxAddCst = LAddCst;
1764 APInt RRangeLow = -RAddCst->getValue();
1765 APInt RRangeHigh = RRangeLow + LHSCst->getValue();
1766 APInt LRangeLow = -LAddCst->getValue();
1767 APInt LRangeHigh = LRangeLow + LHSCst->getValue();
1768 APInt LowRangeDiff = RRangeLow ^ LRangeLow;
1769 APInt HighRangeDiff = RRangeHigh ^ LRangeHigh;
1770 APInt RangeDiff = LRangeLow.sgt(RRangeLow) ? LRangeLow - RRangeLow
1771 : RRangeLow - LRangeLow;
1773 if (LowRangeDiff.isPowerOf2() && LowRangeDiff == HighRangeDiff &&
1774 RangeDiff.ugt(LHSCst->getValue())) {
1775 Value *MaskCst = ConstantInt::get(LAddCst->getType(), ~DiffCst);
1777 Value *NewAnd = Builder->CreateAnd(LAddOpnd, MaskCst);
1778 Value *NewAdd = Builder->CreateAdd(NewAnd, MaxAddCst);
1779 return (Builder->CreateICmp(LHS->getPredicate(), NewAdd, LHSCst));
1785 // (icmp1 A, B) | (icmp2 A, B) --> (icmp3 A, B)
1786 if (PredicatesFoldable(LHSCC, RHSCC)) {
1787 if (LHS->getOperand(0) == RHS->getOperand(1) &&
1788 LHS->getOperand(1) == RHS->getOperand(0))
1789 LHS->swapOperands();
1790 if (LHS->getOperand(0) == RHS->getOperand(0) &&
1791 LHS->getOperand(1) == RHS->getOperand(1)) {
1792 Value *Op0 = LHS->getOperand(0), *Op1 = LHS->getOperand(1);
1793 unsigned Code = getICmpCode(LHS) | getICmpCode(RHS);
1794 bool isSigned = LHS->isSigned() || RHS->isSigned();
1795 return getNewICmpValue(isSigned, Code, Op0, Op1, Builder);
1799 // handle (roughly):
1800 // (icmp ne (A & B), C) | (icmp ne (A & D), E)
1801 if (Value *V = foldLogOpOfMaskedICmps(LHS, RHS, false, Builder))
1804 Value *Val = LHS->getOperand(0), *Val2 = RHS->getOperand(0);
1805 if (LHS->hasOneUse() || RHS->hasOneUse()) {
1806 // (icmp eq B, 0) | (icmp ult A, B) -> (icmp ule A, B-1)
1807 // (icmp eq B, 0) | (icmp ugt B, A) -> (icmp ule A, B-1)
1808 Value *A = nullptr, *B = nullptr;
1809 if (LHSCC == ICmpInst::ICMP_EQ && LHSCst && LHSCst->isZero()) {
1811 if (RHSCC == ICmpInst::ICMP_ULT && Val == RHS->getOperand(1))
1813 else if (RHSCC == ICmpInst::ICMP_UGT && Val == Val2)
1814 A = RHS->getOperand(1);
1816 // (icmp ult A, B) | (icmp eq B, 0) -> (icmp ule A, B-1)
1817 // (icmp ugt B, A) | (icmp eq B, 0) -> (icmp ule A, B-1)
1818 else if (RHSCC == ICmpInst::ICMP_EQ && RHSCst && RHSCst->isZero()) {
1820 if (LHSCC == ICmpInst::ICMP_ULT && Val2 == LHS->getOperand(1))
1822 else if (LHSCC == ICmpInst::ICMP_UGT && Val2 == Val)
1823 A = LHS->getOperand(1);
1826 return Builder->CreateICmp(
1828 Builder->CreateAdd(B, ConstantInt::getSigned(B->getType(), -1)), A);
1831 // E.g. (icmp slt x, 0) | (icmp sgt x, n) --> icmp ugt x, n
1832 if (Value *V = simplifyRangeCheck(LHS, RHS, /*Inverted=*/true))
1835 // E.g. (icmp sgt x, n) | (icmp slt x, 0) --> icmp ugt x, n
1836 if (Value *V = simplifyRangeCheck(RHS, LHS, /*Inverted=*/true))
1839 // This only handles icmp of constants: (icmp1 A, C1) | (icmp2 B, C2).
1840 if (!LHSCst || !RHSCst) return nullptr;
1842 if (LHSCst == RHSCst && LHSCC == RHSCC) {
1843 // (icmp ne A, 0) | (icmp ne B, 0) --> (icmp ne (A|B), 0)
1844 if (LHSCC == ICmpInst::ICMP_NE && LHSCst->isZero()) {
1845 Value *NewOr = Builder->CreateOr(Val, Val2);
1846 return Builder->CreateICmp(LHSCC, NewOr, LHSCst);
1850 // (icmp ult (X + CA), C1) | (icmp eq X, C2) -> (icmp ule (X + CA), C1)
1851 // iff C2 + CA == C1.
1852 if (LHSCC == ICmpInst::ICMP_ULT && RHSCC == ICmpInst::ICMP_EQ) {
1853 ConstantInt *AddCst;
1854 if (match(Val, m_Add(m_Specific(Val2), m_ConstantInt(AddCst))))
1855 if (RHSCst->getValue() + AddCst->getValue() == LHSCst->getValue())
1856 return Builder->CreateICmpULE(Val, LHSCst);
1859 // From here on, we only handle:
1860 // (icmp1 A, C1) | (icmp2 A, C2) --> something simpler.
1861 if (Val != Val2) return nullptr;
1863 // ICMP_[US][GL]E X, CST is folded to ICMP_[US][GL]T elsewhere.
1864 if (LHSCC == ICmpInst::ICMP_UGE || LHSCC == ICmpInst::ICMP_ULE ||
1865 RHSCC == ICmpInst::ICMP_UGE || RHSCC == ICmpInst::ICMP_ULE ||
1866 LHSCC == ICmpInst::ICMP_SGE || LHSCC == ICmpInst::ICMP_SLE ||
1867 RHSCC == ICmpInst::ICMP_SGE || RHSCC == ICmpInst::ICMP_SLE)
1870 // We can't fold (ugt x, C) | (sgt x, C2).
1871 if (!PredicatesFoldable(LHSCC, RHSCC))
1874 // Ensure that the larger constant is on the RHS.
1876 if (CmpInst::isSigned(LHSCC) ||
1877 (ICmpInst::isEquality(LHSCC) &&
1878 CmpInst::isSigned(RHSCC)))
1879 ShouldSwap = LHSCst->getValue().sgt(RHSCst->getValue());
1881 ShouldSwap = LHSCst->getValue().ugt(RHSCst->getValue());
1884 std::swap(LHS, RHS);
1885 std::swap(LHSCst, RHSCst);
1886 std::swap(LHSCC, RHSCC);
1889 // At this point, we know we have two icmp instructions
1890 // comparing a value against two constants and or'ing the result
1891 // together. Because of the above check, we know that we only have
1892 // ICMP_EQ, ICMP_NE, ICMP_LT, and ICMP_GT here. We also know (from the
1893 // icmp folding check above), that the two constants are not
1895 assert(LHSCst != RHSCst && "Compares not folded above?");
1898 default: llvm_unreachable("Unknown integer condition code!");
1899 case ICmpInst::ICMP_EQ:
1901 default: llvm_unreachable("Unknown integer condition code!");
1902 case ICmpInst::ICMP_EQ:
1903 if (LHS->getOperand(0) == RHS->getOperand(0)) {
1904 // if LHSCst and RHSCst differ only by one bit:
1905 // (A == C1 || A == C2) -> (A & ~(C1 ^ C2)) == C1
1906 assert(LHSCst->getValue().ule(LHSCst->getValue()));
1908 APInt Xor = LHSCst->getValue() ^ RHSCst->getValue();
1909 if (Xor.isPowerOf2()) {
1910 Value *NegCst = Builder->getInt(~Xor);
1911 Value *And = Builder->CreateAnd(LHS->getOperand(0), NegCst);
1912 return Builder->CreateICmp(ICmpInst::ICMP_EQ, And, LHSCst);
1916 if (LHSCst == SubOne(RHSCst)) {
1917 // (X == 13 | X == 14) -> X-13 <u 2
1918 Constant *AddCST = ConstantExpr::getNeg(LHSCst);
1919 Value *Add = Builder->CreateAdd(Val, AddCST, Val->getName()+".off");
1920 AddCST = ConstantExpr::getSub(AddOne(RHSCst), LHSCst);
1921 return Builder->CreateICmpULT(Add, AddCST);
1924 break; // (X == 13 | X == 15) -> no change
1925 case ICmpInst::ICMP_UGT: // (X == 13 | X u> 14) -> no change
1926 case ICmpInst::ICMP_SGT: // (X == 13 | X s> 14) -> no change
1928 case ICmpInst::ICMP_NE: // (X == 13 | X != 15) -> X != 15
1929 case ICmpInst::ICMP_ULT: // (X == 13 | X u< 15) -> X u< 15
1930 case ICmpInst::ICMP_SLT: // (X == 13 | X s< 15) -> X s< 15
1934 case ICmpInst::ICMP_NE:
1936 default: llvm_unreachable("Unknown integer condition code!");
1937 case ICmpInst::ICMP_EQ: // (X != 13 | X == 15) -> X != 13
1938 case ICmpInst::ICMP_UGT: // (X != 13 | X u> 15) -> X != 13
1939 case ICmpInst::ICMP_SGT: // (X != 13 | X s> 15) -> X != 13
1941 case ICmpInst::ICMP_NE: // (X != 13 | X != 15) -> true
1942 case ICmpInst::ICMP_ULT: // (X != 13 | X u< 15) -> true
1943 case ICmpInst::ICMP_SLT: // (X != 13 | X s< 15) -> true
1944 return Builder->getTrue();
1946 case ICmpInst::ICMP_ULT:
1948 default: llvm_unreachable("Unknown integer condition code!");
1949 case ICmpInst::ICMP_EQ: // (X u< 13 | X == 14) -> no change
1951 case ICmpInst::ICMP_UGT: // (X u< 13 | X u> 15) -> (X-13) u> 2
1952 // If RHSCst is [us]MAXINT, it is always false. Not handling
1953 // this can cause overflow.
1954 if (RHSCst->isMaxValue(false))
1956 return InsertRangeTest(Val, LHSCst, AddOne(RHSCst), false, false);
1957 case ICmpInst::ICMP_SGT: // (X u< 13 | X s> 15) -> no change
1959 case ICmpInst::ICMP_NE: // (X u< 13 | X != 15) -> X != 15
1960 case ICmpInst::ICMP_ULT: // (X u< 13 | X u< 15) -> X u< 15
1962 case ICmpInst::ICMP_SLT: // (X u< 13 | X s< 15) -> no change
1966 case ICmpInst::ICMP_SLT:
1968 default: llvm_unreachable("Unknown integer condition code!");
1969 case ICmpInst::ICMP_EQ: // (X s< 13 | X == 14) -> no change
1971 case ICmpInst::ICMP_SGT: // (X s< 13 | X s> 15) -> (X-13) s> 2
1972 // If RHSCst is [us]MAXINT, it is always false. Not handling
1973 // this can cause overflow.
1974 if (RHSCst->isMaxValue(true))
1976 return InsertRangeTest(Val, LHSCst, AddOne(RHSCst), true, false);
1977 case ICmpInst::ICMP_UGT: // (X s< 13 | X u> 15) -> no change
1979 case ICmpInst::ICMP_NE: // (X s< 13 | X != 15) -> X != 15
1980 case ICmpInst::ICMP_SLT: // (X s< 13 | X s< 15) -> X s< 15
1982 case ICmpInst::ICMP_ULT: // (X s< 13 | X u< 15) -> no change
1986 case ICmpInst::ICMP_UGT:
1988 default: llvm_unreachable("Unknown integer condition code!");
1989 case ICmpInst::ICMP_EQ: // (X u> 13 | X == 15) -> X u> 13
1990 case ICmpInst::ICMP_UGT: // (X u> 13 | X u> 15) -> X u> 13
1992 case ICmpInst::ICMP_SGT: // (X u> 13 | X s> 15) -> no change
1994 case ICmpInst::ICMP_NE: // (X u> 13 | X != 15) -> true
1995 case ICmpInst::ICMP_ULT: // (X u> 13 | X u< 15) -> true
1996 return Builder->getTrue();
1997 case ICmpInst::ICMP_SLT: // (X u> 13 | X s< 15) -> no change
2001 case ICmpInst::ICMP_SGT:
2003 default: llvm_unreachable("Unknown integer condition code!");
2004 case ICmpInst::ICMP_EQ: // (X s> 13 | X == 15) -> X > 13
2005 case ICmpInst::ICMP_SGT: // (X s> 13 | X s> 15) -> X > 13
2007 case ICmpInst::ICMP_UGT: // (X s> 13 | X u> 15) -> no change
2009 case ICmpInst::ICMP_NE: // (X s> 13 | X != 15) -> true
2010 case ICmpInst::ICMP_SLT: // (X s> 13 | X s< 15) -> true
2011 return Builder->getTrue();
2012 case ICmpInst::ICMP_ULT: // (X s> 13 | X u< 15) -> no change
2020 /// FoldOrOfFCmps - Optimize (fcmp)|(fcmp). NOTE: Unlike the rest of
2021 /// instcombine, this returns a Value which should already be inserted into the
2023 Value *InstCombiner::FoldOrOfFCmps(FCmpInst *LHS, FCmpInst *RHS) {
2024 if (LHS->getPredicate() == FCmpInst::FCMP_UNO &&
2025 RHS->getPredicate() == FCmpInst::FCMP_UNO &&
2026 LHS->getOperand(0)->getType() == RHS->getOperand(0)->getType()) {
2027 if (ConstantFP *LHSC = dyn_cast<ConstantFP>(LHS->getOperand(1)))
2028 if (ConstantFP *RHSC = dyn_cast<ConstantFP>(RHS->getOperand(1))) {
2029 // If either of the constants are nans, then the whole thing returns
2031 if (LHSC->getValueAPF().isNaN() || RHSC->getValueAPF().isNaN())
2032 return Builder->getTrue();
2034 // Otherwise, no need to compare the two constants, compare the
2036 return Builder->CreateFCmpUNO(LHS->getOperand(0), RHS->getOperand(0));
2039 // Handle vector zeros. This occurs because the canonical form of
2040 // "fcmp uno x,x" is "fcmp uno x, 0".
2041 if (isa<ConstantAggregateZero>(LHS->getOperand(1)) &&
2042 isa<ConstantAggregateZero>(RHS->getOperand(1)))
2043 return Builder->CreateFCmpUNO(LHS->getOperand(0), RHS->getOperand(0));
2048 Value *Op0LHS = LHS->getOperand(0), *Op0RHS = LHS->getOperand(1);
2049 Value *Op1LHS = RHS->getOperand(0), *Op1RHS = RHS->getOperand(1);
2050 FCmpInst::Predicate Op0CC = LHS->getPredicate(), Op1CC = RHS->getPredicate();
2052 if (Op0LHS == Op1RHS && Op0RHS == Op1LHS) {
2053 // Swap RHS operands to match LHS.
2054 Op1CC = FCmpInst::getSwappedPredicate(Op1CC);
2055 std::swap(Op1LHS, Op1RHS);
2057 if (Op0LHS == Op1LHS && Op0RHS == Op1RHS) {
2058 // Simplify (fcmp cc0 x, y) | (fcmp cc1 x, y).
2060 return Builder->CreateFCmp((FCmpInst::Predicate)Op0CC, Op0LHS, Op0RHS);
2061 if (Op0CC == FCmpInst::FCMP_TRUE || Op1CC == FCmpInst::FCMP_TRUE)
2062 return ConstantInt::get(CmpInst::makeCmpResultType(LHS->getType()), 1);
2063 if (Op0CC == FCmpInst::FCMP_FALSE)
2065 if (Op1CC == FCmpInst::FCMP_FALSE)
2069 unsigned Op0Pred = getFCmpCode(Op0CC, Op0Ordered);
2070 unsigned Op1Pred = getFCmpCode(Op1CC, Op1Ordered);
2071 if (Op0Ordered == Op1Ordered) {
2072 // If both are ordered or unordered, return a new fcmp with
2073 // or'ed predicates.
2074 return getFCmpValue(Op0Ordered, Op0Pred|Op1Pred, Op0LHS, Op0RHS, Builder);
2080 /// FoldOrWithConstants - This helper function folds:
2082 /// ((A | B) & C1) | (B & C2)
2088 /// when the XOR of the two constants is "all ones" (-1).
2089 Instruction *InstCombiner::FoldOrWithConstants(BinaryOperator &I, Value *Op,
2090 Value *A, Value *B, Value *C) {
2091 ConstantInt *CI1 = dyn_cast<ConstantInt>(C);
2092 if (!CI1) return nullptr;
2094 Value *V1 = nullptr;
2095 ConstantInt *CI2 = nullptr;
2096 if (!match(Op, m_And(m_Value(V1), m_ConstantInt(CI2)))) return nullptr;
2098 APInt Xor = CI1->getValue() ^ CI2->getValue();
2099 if (!Xor.isAllOnesValue()) return nullptr;
2101 if (V1 == A || V1 == B) {
2102 Value *NewOp = Builder->CreateAnd((V1 == A) ? B : A, CI1);
2103 return BinaryOperator::CreateOr(NewOp, V1);
2109 /// \brief This helper function folds:
2111 /// ((A | B) & C1) ^ (B & C2)
2117 /// when the XOR of the two constants is "all ones" (-1).
2118 Instruction *InstCombiner::FoldXorWithConstants(BinaryOperator &I, Value *Op,
2119 Value *A, Value *B, Value *C) {
2120 ConstantInt *CI1 = dyn_cast<ConstantInt>(C);
2124 Value *V1 = nullptr;
2125 ConstantInt *CI2 = nullptr;
2126 if (!match(Op, m_And(m_Value(V1), m_ConstantInt(CI2))))
2129 APInt Xor = CI1->getValue() ^ CI2->getValue();
2130 if (!Xor.isAllOnesValue())
2133 if (V1 == A || V1 == B) {
2134 Value *NewOp = Builder->CreateAnd(V1 == A ? B : A, CI1);
2135 return BinaryOperator::CreateXor(NewOp, V1);
2141 Instruction *InstCombiner::visitOr(BinaryOperator &I) {
2142 bool Changed = SimplifyAssociativeOrCommutative(I);
2143 Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
2145 if (Value *V = SimplifyVectorOp(I))
2146 return ReplaceInstUsesWith(I, V);
2148 if (Value *V = SimplifyOrInst(Op0, Op1, DL, TLI, DT, AC))
2149 return ReplaceInstUsesWith(I, V);
2151 // (A&B)|(A&C) -> A&(B|C) etc
2152 if (Value *V = SimplifyUsingDistributiveLaws(I))
2153 return ReplaceInstUsesWith(I, V);
2155 // See if we can simplify any instructions used by the instruction whose sole
2156 // purpose is to compute bits we don't care about.
2157 if (SimplifyDemandedInstructionBits(I))
2160 if (Value *V = SimplifyBSwap(I))
2161 return ReplaceInstUsesWith(I, V);
2163 if (ConstantInt *RHS = dyn_cast<ConstantInt>(Op1)) {
2164 ConstantInt *C1 = nullptr; Value *X = nullptr;
2165 // (X & C1) | C2 --> (X | C2) & (C1|C2)
2166 // iff (C1 & C2) == 0.
2167 if (match(Op0, m_And(m_Value(X), m_ConstantInt(C1))) &&
2168 (RHS->getValue() & C1->getValue()) != 0 &&
2170 Value *Or = Builder->CreateOr(X, RHS);
2172 return BinaryOperator::CreateAnd(Or,
2173 Builder->getInt(RHS->getValue() | C1->getValue()));
2176 // (X ^ C1) | C2 --> (X | C2) ^ (C1&~C2)
2177 if (match(Op0, m_Xor(m_Value(X), m_ConstantInt(C1))) &&
2179 Value *Or = Builder->CreateOr(X, RHS);
2181 return BinaryOperator::CreateXor(Or,
2182 Builder->getInt(C1->getValue() & ~RHS->getValue()));
2185 // Try to fold constant and into select arguments.
2186 if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
2187 if (Instruction *R = FoldOpIntoSelect(I, SI))
2190 if (isa<PHINode>(Op0))
2191 if (Instruction *NV = FoldOpIntoPhi(I))
2195 Value *A = nullptr, *B = nullptr;
2196 ConstantInt *C1 = nullptr, *C2 = nullptr;
2198 // (A | B) | C and A | (B | C) -> bswap if possible.
2199 // (A >> B) | (C << D) and (A << B) | (B >> C) -> bswap if possible.
2200 if (match(Op0, m_Or(m_Value(), m_Value())) ||
2201 match(Op1, m_Or(m_Value(), m_Value())) ||
2202 (match(Op0, m_LogicalShift(m_Value(), m_Value())) &&
2203 match(Op1, m_LogicalShift(m_Value(), m_Value())))) {
2204 if (Instruction *BSwap = MatchBSwap(I))
2208 // (X^C)|Y -> (X|Y)^C iff Y&C == 0
2209 if (Op0->hasOneUse() &&
2210 match(Op0, m_Xor(m_Value(A), m_ConstantInt(C1))) &&
2211 MaskedValueIsZero(Op1, C1->getValue(), 0, &I)) {
2212 Value *NOr = Builder->CreateOr(A, Op1);
2214 return BinaryOperator::CreateXor(NOr, C1);
2217 // Y|(X^C) -> (X|Y)^C iff Y&C == 0
2218 if (Op1->hasOneUse() &&
2219 match(Op1, m_Xor(m_Value(A), m_ConstantInt(C1))) &&
2220 MaskedValueIsZero(Op0, C1->getValue(), 0, &I)) {
2221 Value *NOr = Builder->CreateOr(A, Op0);
2223 return BinaryOperator::CreateXor(NOr, C1);
2226 // ((~A & B) | A) -> (A | B)
2227 if (match(Op0, m_And(m_Not(m_Value(A)), m_Value(B))) &&
2228 match(Op1, m_Specific(A)))
2229 return BinaryOperator::CreateOr(A, B);
2231 // ((A & B) | ~A) -> (~A | B)
2232 if (match(Op0, m_And(m_Value(A), m_Value(B))) &&
2233 match(Op1, m_Not(m_Specific(A))))
2234 return BinaryOperator::CreateOr(Builder->CreateNot(A), B);
2236 // (A & (~B)) | (A ^ B) -> (A ^ B)
2237 if (match(Op0, m_And(m_Value(A), m_Not(m_Value(B)))) &&
2238 match(Op1, m_Xor(m_Specific(A), m_Specific(B))))
2239 return BinaryOperator::CreateXor(A, B);
2241 // (A ^ B) | ( A & (~B)) -> (A ^ B)
2242 if (match(Op0, m_Xor(m_Value(A), m_Value(B))) &&
2243 match(Op1, m_And(m_Specific(A), m_Not(m_Specific(B)))))
2244 return BinaryOperator::CreateXor(A, B);
2247 Value *C = nullptr, *D = nullptr;
2248 if (match(Op0, m_And(m_Value(A), m_Value(C))) &&
2249 match(Op1, m_And(m_Value(B), m_Value(D)))) {
2250 Value *V1 = nullptr, *V2 = nullptr;
2251 C1 = dyn_cast<ConstantInt>(C);
2252 C2 = dyn_cast<ConstantInt>(D);
2253 if (C1 && C2) { // (A & C1)|(B & C2)
2254 if ((C1->getValue() & C2->getValue()) == 0) {
2255 // ((V | N) & C1) | (V & C2) --> (V|N) & (C1|C2)
2256 // iff (C1&C2) == 0 and (N&~C1) == 0
2257 if (match(A, m_Or(m_Value(V1), m_Value(V2))) &&
2259 MaskedValueIsZero(V2, ~C1->getValue(), 0, &I)) || // (V|N)
2261 MaskedValueIsZero(V1, ~C1->getValue(), 0, &I)))) // (N|V)
2262 return BinaryOperator::CreateAnd(A,
2263 Builder->getInt(C1->getValue()|C2->getValue()));
2264 // Or commutes, try both ways.
2265 if (match(B, m_Or(m_Value(V1), m_Value(V2))) &&
2267 MaskedValueIsZero(V2, ~C2->getValue(), 0, &I)) || // (V|N)
2269 MaskedValueIsZero(V1, ~C2->getValue(), 0, &I)))) // (N|V)
2270 return BinaryOperator::CreateAnd(B,
2271 Builder->getInt(C1->getValue()|C2->getValue()));
2273 // ((V|C3)&C1) | ((V|C4)&C2) --> (V|C3|C4)&(C1|C2)
2274 // iff (C1&C2) == 0 and (C3&~C1) == 0 and (C4&~C2) == 0.
2275 ConstantInt *C3 = nullptr, *C4 = nullptr;
2276 if (match(A, m_Or(m_Value(V1), m_ConstantInt(C3))) &&
2277 (C3->getValue() & ~C1->getValue()) == 0 &&
2278 match(B, m_Or(m_Specific(V1), m_ConstantInt(C4))) &&
2279 (C4->getValue() & ~C2->getValue()) == 0) {
2280 V2 = Builder->CreateOr(V1, ConstantExpr::getOr(C3, C4), "bitfield");
2281 return BinaryOperator::CreateAnd(V2,
2282 Builder->getInt(C1->getValue()|C2->getValue()));
2287 // (A & (C0?-1:0)) | (B & ~(C0?-1:0)) -> C0 ? A : B, and commuted variants.
2288 // Don't do this for vector select idioms, the code generator doesn't handle
2290 if (!I.getType()->isVectorTy()) {
2291 if (Instruction *Match = MatchSelectFromAndOr(A, B, C, D))
2293 if (Instruction *Match = MatchSelectFromAndOr(B, A, D, C))
2295 if (Instruction *Match = MatchSelectFromAndOr(C, B, A, D))
2297 if (Instruction *Match = MatchSelectFromAndOr(D, A, B, C))
2301 // ((A&~B)|(~A&B)) -> A^B
2302 if ((match(C, m_Not(m_Specific(D))) &&
2303 match(B, m_Not(m_Specific(A)))))
2304 return BinaryOperator::CreateXor(A, D);
2305 // ((~B&A)|(~A&B)) -> A^B
2306 if ((match(A, m_Not(m_Specific(D))) &&
2307 match(B, m_Not(m_Specific(C)))))
2308 return BinaryOperator::CreateXor(C, D);
2309 // ((A&~B)|(B&~A)) -> A^B
2310 if ((match(C, m_Not(m_Specific(B))) &&
2311 match(D, m_Not(m_Specific(A)))))
2312 return BinaryOperator::CreateXor(A, B);
2313 // ((~B&A)|(B&~A)) -> A^B
2314 if ((match(A, m_Not(m_Specific(B))) &&
2315 match(D, m_Not(m_Specific(C)))))
2316 return BinaryOperator::CreateXor(C, B);
2318 // ((A|B)&1)|(B&-2) -> (A&1) | B
2319 if (match(A, m_Or(m_Value(V1), m_Specific(B))) ||
2320 match(A, m_Or(m_Specific(B), m_Value(V1)))) {
2321 Instruction *Ret = FoldOrWithConstants(I, Op1, V1, B, C);
2322 if (Ret) return Ret;
2324 // (B&-2)|((A|B)&1) -> (A&1) | B
2325 if (match(B, m_Or(m_Specific(A), m_Value(V1))) ||
2326 match(B, m_Or(m_Value(V1), m_Specific(A)))) {
2327 Instruction *Ret = FoldOrWithConstants(I, Op0, A, V1, D);
2328 if (Ret) return Ret;
2330 // ((A^B)&1)|(B&-2) -> (A&1) ^ B
2331 if (match(A, m_Xor(m_Value(V1), m_Specific(B))) ||
2332 match(A, m_Xor(m_Specific(B), m_Value(V1)))) {
2333 Instruction *Ret = FoldXorWithConstants(I, Op1, V1, B, C);
2334 if (Ret) return Ret;
2336 // (B&-2)|((A^B)&1) -> (A&1) ^ B
2337 if (match(B, m_Xor(m_Specific(A), m_Value(V1))) ||
2338 match(B, m_Xor(m_Value(V1), m_Specific(A)))) {
2339 Instruction *Ret = FoldXorWithConstants(I, Op0, A, V1, D);
2340 if (Ret) return Ret;
2344 // (A ^ B) | ((B ^ C) ^ A) -> (A ^ B) | C
2345 if (match(Op0, m_Xor(m_Value(A), m_Value(B))))
2346 if (match(Op1, m_Xor(m_Xor(m_Specific(B), m_Value(C)), m_Specific(A))))
2347 if (Op1->hasOneUse() || cast<BinaryOperator>(Op1)->hasOneUse())
2348 return BinaryOperator::CreateOr(Op0, C);
2350 // ((A ^ C) ^ B) | (B ^ A) -> (B ^ A) | C
2351 if (match(Op0, m_Xor(m_Xor(m_Value(A), m_Value(C)), m_Value(B))))
2352 if (match(Op1, m_Xor(m_Specific(B), m_Specific(A))))
2353 if (Op0->hasOneUse() || cast<BinaryOperator>(Op0)->hasOneUse())
2354 return BinaryOperator::CreateOr(Op1, C);
2356 // ((B | C) & A) | B -> B | (A & C)
2357 if (match(Op0, m_And(m_Or(m_Specific(Op1), m_Value(C)), m_Value(A))))
2358 return BinaryOperator::CreateOr(Op1, Builder->CreateAnd(A, C));
2360 // (~A | ~B) == (~(A & B)) - De Morgan's Law
2361 if (Value *Op0NotVal = dyn_castNotVal(Op0))
2362 if (Value *Op1NotVal = dyn_castNotVal(Op1))
2363 if (Op0->hasOneUse() && Op1->hasOneUse()) {
2364 Value *And = Builder->CreateAnd(Op0NotVal, Op1NotVal,
2365 I.getName()+".demorgan");
2366 return BinaryOperator::CreateNot(And);
2369 // Canonicalize xor to the RHS.
2370 bool SwappedForXor = false;
2371 if (match(Op0, m_Xor(m_Value(), m_Value()))) {
2372 std::swap(Op0, Op1);
2373 SwappedForXor = true;
2376 // A | ( A ^ B) -> A | B
2377 // A | (~A ^ B) -> A | ~B
2378 // (A & B) | (A ^ B)
2379 if (match(Op1, m_Xor(m_Value(A), m_Value(B)))) {
2380 if (Op0 == A || Op0 == B)
2381 return BinaryOperator::CreateOr(A, B);
2383 if (match(Op0, m_And(m_Specific(A), m_Specific(B))) ||
2384 match(Op0, m_And(m_Specific(B), m_Specific(A))))
2385 return BinaryOperator::CreateOr(A, B);
2387 if (Op1->hasOneUse() && match(A, m_Not(m_Specific(Op0)))) {
2388 Value *Not = Builder->CreateNot(B, B->getName()+".not");
2389 return BinaryOperator::CreateOr(Not, Op0);
2391 if (Op1->hasOneUse() && match(B, m_Not(m_Specific(Op0)))) {
2392 Value *Not = Builder->CreateNot(A, A->getName()+".not");
2393 return BinaryOperator::CreateOr(Not, Op0);
2397 // A | ~(A | B) -> A | ~B
2398 // A | ~(A ^ B) -> A | ~B
2399 if (match(Op1, m_Not(m_Value(A))))
2400 if (BinaryOperator *B = dyn_cast<BinaryOperator>(A))
2401 if ((Op0 == B->getOperand(0) || Op0 == B->getOperand(1)) &&
2402 Op1->hasOneUse() && (B->getOpcode() == Instruction::Or ||
2403 B->getOpcode() == Instruction::Xor)) {
2404 Value *NotOp = Op0 == B->getOperand(0) ? B->getOperand(1) :
2406 Value *Not = Builder->CreateNot(NotOp, NotOp->getName()+".not");
2407 return BinaryOperator::CreateOr(Not, Op0);
2410 // (A & B) | ((~A) ^ B) -> (~A ^ B)
2411 if (match(Op0, m_And(m_Value(A), m_Value(B))) &&
2412 match(Op1, m_Xor(m_Not(m_Specific(A)), m_Specific(B))))
2413 return BinaryOperator::CreateXor(Builder->CreateNot(A), B);
2415 // ((~A) ^ B) | (A & B) -> (~A ^ B)
2416 if (match(Op0, m_Xor(m_Not(m_Value(A)), m_Value(B))) &&
2417 match(Op1, m_And(m_Specific(A), m_Specific(B))))
2418 return BinaryOperator::CreateXor(Builder->CreateNot(A), B);
2421 std::swap(Op0, Op1);
2424 ICmpInst *LHS = dyn_cast<ICmpInst>(Op0);
2425 ICmpInst *RHS = dyn_cast<ICmpInst>(Op1);
2427 if (Value *Res = FoldOrOfICmps(LHS, RHS, &I))
2428 return ReplaceInstUsesWith(I, Res);
2430 // TODO: Make this recursive; it's a little tricky because an arbitrary
2431 // number of 'or' instructions might have to be created.
2433 if (LHS && match(Op1, m_OneUse(m_Or(m_Value(X), m_Value(Y))))) {
2434 if (auto *Cmp = dyn_cast<ICmpInst>(X))
2435 if (Value *Res = FoldOrOfICmps(LHS, Cmp, &I))
2436 return ReplaceInstUsesWith(I, Builder->CreateOr(Res, Y));
2437 if (auto *Cmp = dyn_cast<ICmpInst>(Y))
2438 if (Value *Res = FoldOrOfICmps(LHS, Cmp, &I))
2439 return ReplaceInstUsesWith(I, Builder->CreateOr(Res, X));
2441 if (RHS && match(Op0, m_OneUse(m_Or(m_Value(X), m_Value(Y))))) {
2442 if (auto *Cmp = dyn_cast<ICmpInst>(X))
2443 if (Value *Res = FoldOrOfICmps(Cmp, RHS, &I))
2444 return ReplaceInstUsesWith(I, Builder->CreateOr(Res, Y));
2445 if (auto *Cmp = dyn_cast<ICmpInst>(Y))
2446 if (Value *Res = FoldOrOfICmps(Cmp, RHS, &I))
2447 return ReplaceInstUsesWith(I, Builder->CreateOr(Res, X));
2451 // (fcmp uno x, c) | (fcmp uno y, c) -> (fcmp uno x, y)
2452 if (FCmpInst *LHS = dyn_cast<FCmpInst>(I.getOperand(0)))
2453 if (FCmpInst *RHS = dyn_cast<FCmpInst>(I.getOperand(1)))
2454 if (Value *Res = FoldOrOfFCmps(LHS, RHS))
2455 return ReplaceInstUsesWith(I, Res);
2457 // fold (or (cast A), (cast B)) -> (cast (or A, B))
2458 if (CastInst *Op0C = dyn_cast<CastInst>(Op0)) {
2459 CastInst *Op1C = dyn_cast<CastInst>(Op1);
2460 if (Op1C && Op0C->getOpcode() == Op1C->getOpcode()) {// same cast kind ?
2461 Type *SrcTy = Op0C->getOperand(0)->getType();
2462 if (SrcTy == Op1C->getOperand(0)->getType() &&
2463 SrcTy->isIntOrIntVectorTy()) {
2464 Value *Op0COp = Op0C->getOperand(0), *Op1COp = Op1C->getOperand(0);
2466 if ((!isa<ICmpInst>(Op0COp) || !isa<ICmpInst>(Op1COp)) &&
2467 // Only do this if the casts both really cause code to be
2469 ShouldOptimizeCast(Op0C->getOpcode(), Op0COp, I.getType()) &&
2470 ShouldOptimizeCast(Op1C->getOpcode(), Op1COp, I.getType())) {
2471 Value *NewOp = Builder->CreateOr(Op0COp, Op1COp, I.getName());
2472 return CastInst::Create(Op0C->getOpcode(), NewOp, I.getType());
2475 // If this is or(cast(icmp), cast(icmp)), try to fold this even if the
2476 // cast is otherwise not optimizable. This happens for vector sexts.
2477 if (ICmpInst *RHS = dyn_cast<ICmpInst>(Op1COp))
2478 if (ICmpInst *LHS = dyn_cast<ICmpInst>(Op0COp))
2479 if (Value *Res = FoldOrOfICmps(LHS, RHS, &I))
2480 return CastInst::Create(Op0C->getOpcode(), Res, I.getType());
2482 // If this is or(cast(fcmp), cast(fcmp)), try to fold this even if the
2483 // cast is otherwise not optimizable. This happens for vector sexts.
2484 if (FCmpInst *RHS = dyn_cast<FCmpInst>(Op1COp))
2485 if (FCmpInst *LHS = dyn_cast<FCmpInst>(Op0COp))
2486 if (Value *Res = FoldOrOfFCmps(LHS, RHS))
2487 return CastInst::Create(Op0C->getOpcode(), Res, I.getType());
2492 // or(sext(A), B) -> A ? -1 : B where A is an i1
2493 // or(A, sext(B)) -> B ? -1 : A where B is an i1
2494 if (match(Op0, m_SExt(m_Value(A))) && A->getType()->isIntegerTy(1))
2495 return SelectInst::Create(A, ConstantInt::getSigned(I.getType(), -1), Op1);
2496 if (match(Op1, m_SExt(m_Value(A))) && A->getType()->isIntegerTy(1))
2497 return SelectInst::Create(A, ConstantInt::getSigned(I.getType(), -1), Op0);
2499 // Note: If we've gotten to the point of visiting the outer OR, then the
2500 // inner one couldn't be simplified. If it was a constant, then it won't
2501 // be simplified by a later pass either, so we try swapping the inner/outer
2502 // ORs in the hopes that we'll be able to simplify it this way.
2503 // (X|C) | V --> (X|V) | C
2504 if (Op0->hasOneUse() && !isa<ConstantInt>(Op1) &&
2505 match(Op0, m_Or(m_Value(A), m_ConstantInt(C1)))) {
2506 Value *Inner = Builder->CreateOr(A, Op1);
2507 Inner->takeName(Op0);
2508 return BinaryOperator::CreateOr(Inner, C1);
2511 // Change (or (bool?A:B),(bool?C:D)) --> (bool?(or A,C):(or B,D))
2512 // Since this OR statement hasn't been optimized further yet, we hope
2513 // that this transformation will allow the new ORs to be optimized.
2515 Value *X = nullptr, *Y = nullptr;
2516 if (Op0->hasOneUse() && Op1->hasOneUse() &&
2517 match(Op0, m_Select(m_Value(X), m_Value(A), m_Value(B))) &&
2518 match(Op1, m_Select(m_Value(Y), m_Value(C), m_Value(D))) && X == Y) {
2519 Value *orTrue = Builder->CreateOr(A, C);
2520 Value *orFalse = Builder->CreateOr(B, D);
2521 return SelectInst::Create(X, orTrue, orFalse);
2525 return Changed ? &I : nullptr;
2528 Instruction *InstCombiner::visitXor(BinaryOperator &I) {
2529 bool Changed = SimplifyAssociativeOrCommutative(I);
2530 Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
2532 if (Value *V = SimplifyVectorOp(I))
2533 return ReplaceInstUsesWith(I, V);
2535 if (Value *V = SimplifyXorInst(Op0, Op1, DL, TLI, DT, AC))
2536 return ReplaceInstUsesWith(I, V);
2538 // (A&B)^(A&C) -> A&(B^C) etc
2539 if (Value *V = SimplifyUsingDistributiveLaws(I))
2540 return ReplaceInstUsesWith(I, V);
2542 // See if we can simplify any instructions used by the instruction whose sole
2543 // purpose is to compute bits we don't care about.
2544 if (SimplifyDemandedInstructionBits(I))
2547 if (Value *V = SimplifyBSwap(I))
2548 return ReplaceInstUsesWith(I, V);
2550 // Is this a ~ operation?
2551 if (Value *NotOp = dyn_castNotVal(&I)) {
2552 if (BinaryOperator *Op0I = dyn_cast<BinaryOperator>(NotOp)) {
2553 if (Op0I->getOpcode() == Instruction::And ||
2554 Op0I->getOpcode() == Instruction::Or) {
2555 // ~(~X & Y) --> (X | ~Y) - De Morgan's Law
2556 // ~(~X | Y) === (X & ~Y) - De Morgan's Law
2557 if (dyn_castNotVal(Op0I->getOperand(1)))
2558 Op0I->swapOperands();
2559 if (Value *Op0NotVal = dyn_castNotVal(Op0I->getOperand(0))) {
2561 Builder->CreateNot(Op0I->getOperand(1),
2562 Op0I->getOperand(1)->getName()+".not");
2563 if (Op0I->getOpcode() == Instruction::And)
2564 return BinaryOperator::CreateOr(Op0NotVal, NotY);
2565 return BinaryOperator::CreateAnd(Op0NotVal, NotY);
2568 // ~(X & Y) --> (~X | ~Y) - De Morgan's Law
2569 // ~(X | Y) === (~X & ~Y) - De Morgan's Law
2570 if (IsFreeToInvert(Op0I->getOperand(0),
2571 Op0I->getOperand(0)->hasOneUse()) &&
2572 IsFreeToInvert(Op0I->getOperand(1),
2573 Op0I->getOperand(1)->hasOneUse())) {
2575 Builder->CreateNot(Op0I->getOperand(0), "notlhs");
2577 Builder->CreateNot(Op0I->getOperand(1), "notrhs");
2578 if (Op0I->getOpcode() == Instruction::And)
2579 return BinaryOperator::CreateOr(NotX, NotY);
2580 return BinaryOperator::CreateAnd(NotX, NotY);
2583 } else if (Op0I->getOpcode() == Instruction::AShr) {
2584 // ~(~X >>s Y) --> (X >>s Y)
2585 if (Value *Op0NotVal = dyn_castNotVal(Op0I->getOperand(0)))
2586 return BinaryOperator::CreateAShr(Op0NotVal, Op0I->getOperand(1));
2591 if (Constant *RHS = dyn_cast<Constant>(Op1)) {
2592 if (RHS->isAllOnesValue() && Op0->hasOneUse())
2593 // xor (cmp A, B), true = not (cmp A, B) = !cmp A, B
2594 if (CmpInst *CI = dyn_cast<CmpInst>(Op0))
2595 return CmpInst::Create(CI->getOpcode(),
2596 CI->getInversePredicate(),
2597 CI->getOperand(0), CI->getOperand(1));
2600 if (ConstantInt *RHS = dyn_cast<ConstantInt>(Op1)) {
2601 // fold (xor(zext(cmp)), 1) and (xor(sext(cmp)), -1) to ext(!cmp).
2602 if (CastInst *Op0C = dyn_cast<CastInst>(Op0)) {
2603 if (CmpInst *CI = dyn_cast<CmpInst>(Op0C->getOperand(0))) {
2604 if (CI->hasOneUse() && Op0C->hasOneUse()) {
2605 Instruction::CastOps Opcode = Op0C->getOpcode();
2606 if ((Opcode == Instruction::ZExt || Opcode == Instruction::SExt) &&
2607 (RHS == ConstantExpr::getCast(Opcode, Builder->getTrue(),
2608 Op0C->getDestTy()))) {
2609 CI->setPredicate(CI->getInversePredicate());
2610 return CastInst::Create(Opcode, CI, Op0C->getType());
2616 if (BinaryOperator *Op0I = dyn_cast<BinaryOperator>(Op0)) {
2617 // ~(c-X) == X-c-1 == X+(-c-1)
2618 if (Op0I->getOpcode() == Instruction::Sub && RHS->isAllOnesValue())
2619 if (Constant *Op0I0C = dyn_cast<Constant>(Op0I->getOperand(0))) {
2620 Constant *NegOp0I0C = ConstantExpr::getNeg(Op0I0C);
2621 Constant *ConstantRHS = ConstantExpr::getSub(NegOp0I0C,
2622 ConstantInt::get(I.getType(), 1));
2623 return BinaryOperator::CreateAdd(Op0I->getOperand(1), ConstantRHS);
2626 if (ConstantInt *Op0CI = dyn_cast<ConstantInt>(Op0I->getOperand(1))) {
2627 if (Op0I->getOpcode() == Instruction::Add) {
2628 // ~(X-c) --> (-c-1)-X
2629 if (RHS->isAllOnesValue()) {
2630 Constant *NegOp0CI = ConstantExpr::getNeg(Op0CI);
2631 return BinaryOperator::CreateSub(
2632 ConstantExpr::getSub(NegOp0CI,
2633 ConstantInt::get(I.getType(), 1)),
2634 Op0I->getOperand(0));
2635 } else if (RHS->getValue().isSignBit()) {
2636 // (X + C) ^ signbit -> (X + C + signbit)
2637 Constant *C = Builder->getInt(RHS->getValue() + Op0CI->getValue());
2638 return BinaryOperator::CreateAdd(Op0I->getOperand(0), C);
2641 } else if (Op0I->getOpcode() == Instruction::Or) {
2642 // (X|C1)^C2 -> X^(C1|C2) iff X&~C1 == 0
2643 if (MaskedValueIsZero(Op0I->getOperand(0), Op0CI->getValue(),
2645 Constant *NewRHS = ConstantExpr::getOr(Op0CI, RHS);
2646 // Anything in both C1 and C2 is known to be zero, remove it from
2648 Constant *CommonBits = ConstantExpr::getAnd(Op0CI, RHS);
2649 NewRHS = ConstantExpr::getAnd(NewRHS,
2650 ConstantExpr::getNot(CommonBits));
2652 I.setOperand(0, Op0I->getOperand(0));
2653 I.setOperand(1, NewRHS);
2656 } else if (Op0I->getOpcode() == Instruction::LShr) {
2657 // ((X^C1) >> C2) ^ C3 -> (X>>C2) ^ ((C1>>C2)^C3)
2661 if (Op0I->hasOneUse() &&
2662 (E1 = dyn_cast<BinaryOperator>(Op0I->getOperand(0))) &&
2663 E1->getOpcode() == Instruction::Xor &&
2664 (C1 = dyn_cast<ConstantInt>(E1->getOperand(1)))) {
2665 // fold (C1 >> C2) ^ C3
2666 ConstantInt *C2 = Op0CI, *C3 = RHS;
2667 APInt FoldConst = C1->getValue().lshr(C2->getValue());
2668 FoldConst ^= C3->getValue();
2669 // Prepare the two operands.
2670 Value *Opnd0 = Builder->CreateLShr(E1->getOperand(0), C2);
2671 Opnd0->takeName(Op0I);
2672 cast<Instruction>(Opnd0)->setDebugLoc(I.getDebugLoc());
2673 Value *FoldVal = ConstantInt::get(Opnd0->getType(), FoldConst);
2675 return BinaryOperator::CreateXor(Opnd0, FoldVal);
2681 // Try to fold constant and into select arguments.
2682 if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
2683 if (Instruction *R = FoldOpIntoSelect(I, SI))
2685 if (isa<PHINode>(Op0))
2686 if (Instruction *NV = FoldOpIntoPhi(I))
2690 BinaryOperator *Op1I = dyn_cast<BinaryOperator>(Op1);
2693 if (match(Op1I, m_Or(m_Value(A), m_Value(B)))) {
2694 if (A == Op0) { // B^(B|A) == (A|B)^B
2695 Op1I->swapOperands();
2697 std::swap(Op0, Op1);
2698 } else if (B == Op0) { // B^(A|B) == (A|B)^B
2699 I.swapOperands(); // Simplified below.
2700 std::swap(Op0, Op1);
2702 } else if (match(Op1I, m_And(m_Value(A), m_Value(B))) &&
2704 if (A == Op0) { // A^(A&B) -> A^(B&A)
2705 Op1I->swapOperands();
2708 if (B == Op0) { // A^(B&A) -> (B&A)^A
2709 I.swapOperands(); // Simplified below.
2710 std::swap(Op0, Op1);
2715 BinaryOperator *Op0I = dyn_cast<BinaryOperator>(Op0);
2718 if (match(Op0I, m_Or(m_Value(A), m_Value(B))) &&
2719 Op0I->hasOneUse()) {
2720 if (A == Op1) // (B|A)^B == (A|B)^B
2722 if (B == Op1) // (A|B)^B == A & ~B
2723 return BinaryOperator::CreateAnd(A, Builder->CreateNot(Op1));
2724 } else if (match(Op0I, m_And(m_Value(A), m_Value(B))) &&
2726 if (A == Op1) // (A&B)^A -> (B&A)^A
2728 if (B == Op1 && // (B&A)^A == ~B & A
2729 !isa<ConstantInt>(Op1)) { // Canonical form is (B&C)^C
2730 return BinaryOperator::CreateAnd(Builder->CreateNot(A), Op1);
2736 Value *A, *B, *C, *D;
2737 // (A & B)^(A | B) -> A ^ B
2738 if (match(Op0I, m_And(m_Value(A), m_Value(B))) &&
2739 match(Op1I, m_Or(m_Value(C), m_Value(D)))) {
2740 if ((A == C && B == D) || (A == D && B == C))
2741 return BinaryOperator::CreateXor(A, B);
2743 // (A | B)^(A & B) -> A ^ B
2744 if (match(Op0I, m_Or(m_Value(A), m_Value(B))) &&
2745 match(Op1I, m_And(m_Value(C), m_Value(D)))) {
2746 if ((A == C && B == D) || (A == D && B == C))
2747 return BinaryOperator::CreateXor(A, B);
2749 // (A | ~B) ^ (~A | B) -> A ^ B
2750 if (match(Op0I, m_Or(m_Value(A), m_Not(m_Value(B)))) &&
2751 match(Op1I, m_Or(m_Not(m_Specific(A)), m_Specific(B)))) {
2752 return BinaryOperator::CreateXor(A, B);
2754 // (~A | B) ^ (A | ~B) -> A ^ B
2755 if (match(Op0I, m_Or(m_Not(m_Value(A)), m_Value(B))) &&
2756 match(Op1I, m_Or(m_Specific(A), m_Not(m_Specific(B))))) {
2757 return BinaryOperator::CreateXor(A, B);
2759 // (A & ~B) ^ (~A & B) -> A ^ B
2760 if (match(Op0I, m_And(m_Value(A), m_Not(m_Value(B)))) &&
2761 match(Op1I, m_And(m_Not(m_Specific(A)), m_Specific(B)))) {
2762 return BinaryOperator::CreateXor(A, B);
2764 // (~A & B) ^ (A & ~B) -> A ^ B
2765 if (match(Op0I, m_And(m_Not(m_Value(A)), m_Value(B))) &&
2766 match(Op1I, m_And(m_Specific(A), m_Not(m_Specific(B))))) {
2767 return BinaryOperator::CreateXor(A, B);
2769 // (A ^ C)^(A | B) -> ((~A) & B) ^ C
2770 if (match(Op0I, m_Xor(m_Value(D), m_Value(C))) &&
2771 match(Op1I, m_Or(m_Value(A), m_Value(B)))) {
2773 return BinaryOperator::CreateXor(
2774 Builder->CreateAnd(Builder->CreateNot(A), B), C);
2776 return BinaryOperator::CreateXor(
2777 Builder->CreateAnd(Builder->CreateNot(B), A), C);
2779 // (A | B)^(A ^ C) -> ((~A) & B) ^ C
2780 if (match(Op0I, m_Or(m_Value(A), m_Value(B))) &&
2781 match(Op1I, m_Xor(m_Value(D), m_Value(C)))) {
2783 return BinaryOperator::CreateXor(
2784 Builder->CreateAnd(Builder->CreateNot(A), B), C);
2786 return BinaryOperator::CreateXor(
2787 Builder->CreateAnd(Builder->CreateNot(B), A), C);
2789 // (A & B) ^ (A ^ B) -> (A | B)
2790 if (match(Op0I, m_And(m_Value(A), m_Value(B))) &&
2791 match(Op1I, m_Xor(m_Specific(A), m_Specific(B))))
2792 return BinaryOperator::CreateOr(A, B);
2793 // (A ^ B) ^ (A & B) -> (A | B)
2794 if (match(Op0I, m_Xor(m_Value(A), m_Value(B))) &&
2795 match(Op1I, m_And(m_Specific(A), m_Specific(B))))
2796 return BinaryOperator::CreateOr(A, B);
2799 Value *A = nullptr, *B = nullptr;
2800 // (A & ~B) ^ (~A) -> ~(A & B)
2801 if (match(Op0, m_And(m_Value(A), m_Not(m_Value(B)))) &&
2802 match(Op1, m_Not(m_Specific(A))))
2803 return BinaryOperator::CreateNot(Builder->CreateAnd(A, B));
2805 // (icmp1 A, B) ^ (icmp2 A, B) --> (icmp3 A, B)
2806 if (ICmpInst *RHS = dyn_cast<ICmpInst>(I.getOperand(1)))
2807 if (ICmpInst *LHS = dyn_cast<ICmpInst>(I.getOperand(0)))
2808 if (PredicatesFoldable(LHS->getPredicate(), RHS->getPredicate())) {
2809 if (LHS->getOperand(0) == RHS->getOperand(1) &&
2810 LHS->getOperand(1) == RHS->getOperand(0))
2811 LHS->swapOperands();
2812 if (LHS->getOperand(0) == RHS->getOperand(0) &&
2813 LHS->getOperand(1) == RHS->getOperand(1)) {
2814 Value *Op0 = LHS->getOperand(0), *Op1 = LHS->getOperand(1);
2815 unsigned Code = getICmpCode(LHS) ^ getICmpCode(RHS);
2816 bool isSigned = LHS->isSigned() || RHS->isSigned();
2817 return ReplaceInstUsesWith(I,
2818 getNewICmpValue(isSigned, Code, Op0, Op1,
2823 // fold (xor (cast A), (cast B)) -> (cast (xor A, B))
2824 if (CastInst *Op0C = dyn_cast<CastInst>(Op0)) {
2825 if (CastInst *Op1C = dyn_cast<CastInst>(Op1))
2826 if (Op0C->getOpcode() == Op1C->getOpcode()) { // same cast kind?
2827 Type *SrcTy = Op0C->getOperand(0)->getType();
2828 if (SrcTy == Op1C->getOperand(0)->getType() && SrcTy->isIntegerTy() &&
2829 // Only do this if the casts both really cause code to be generated.
2830 ShouldOptimizeCast(Op0C->getOpcode(), Op0C->getOperand(0),
2832 ShouldOptimizeCast(Op1C->getOpcode(), Op1C->getOperand(0),
2834 Value *NewOp = Builder->CreateXor(Op0C->getOperand(0),
2835 Op1C->getOperand(0), I.getName());
2836 return CastInst::Create(Op0C->getOpcode(), NewOp, I.getType());
2841 return Changed ? &I : nullptr;