1 //===- ConstantFold.cpp - LLVM constant folder ----------------------------===//
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 folding of constants for LLVM. This implements the
11 // (internal) ConstantFold.h interface, which is used by the
12 // ConstantExpr::get* methods to automatically fold constants when possible.
14 // The current constant folding implementation is implemented in two pieces: the
15 // pieces that don't need TargetData, and the pieces that do. This is to avoid
16 // a dependence in VMCore on Target.
18 //===----------------------------------------------------------------------===//
20 #include "ConstantFold.h"
21 #include "llvm/Constants.h"
22 #include "llvm/Instructions.h"
23 #include "llvm/DerivedTypes.h"
24 #include "llvm/Function.h"
25 #include "llvm/GlobalAlias.h"
26 #include "llvm/GlobalVariable.h"
27 #include "llvm/LLVMContext.h"
28 #include "llvm/ADT/SmallVector.h"
29 #include "llvm/Support/Compiler.h"
30 #include "llvm/Support/ErrorHandling.h"
31 #include "llvm/Support/GetElementPtrTypeIterator.h"
32 #include "llvm/Support/ManagedStatic.h"
33 #include "llvm/Support/MathExtras.h"
37 //===----------------------------------------------------------------------===//
38 // ConstantFold*Instruction Implementations
39 //===----------------------------------------------------------------------===//
41 /// BitCastConstantVector - Convert the specified ConstantVector node to the
42 /// specified vector type. At this point, we know that the elements of the
43 /// input vector constant are all simple integer or FP values.
44 static Constant *BitCastConstantVector(LLVMContext &Context, ConstantVector *CV,
45 const VectorType *DstTy) {
46 // If this cast changes element count then we can't handle it here:
47 // doing so requires endianness information. This should be handled by
48 // Analysis/ConstantFolding.cpp
49 unsigned NumElts = DstTy->getNumElements();
50 if (NumElts != CV->getNumOperands())
53 // Check to verify that all elements of the input are simple.
54 for (unsigned i = 0; i != NumElts; ++i) {
55 if (!isa<ConstantInt>(CV->getOperand(i)) &&
56 !isa<ConstantFP>(CV->getOperand(i)))
60 // Bitcast each element now.
61 std::vector<Constant*> Result;
62 const Type *DstEltTy = DstTy->getElementType();
63 for (unsigned i = 0; i != NumElts; ++i)
64 Result.push_back(ConstantExpr::getBitCast(CV->getOperand(i),
66 return ConstantVector::get(Result);
69 /// This function determines which opcode to use to fold two constant cast
70 /// expressions together. It uses CastInst::isEliminableCastPair to determine
71 /// the opcode. Consequently its just a wrapper around that function.
72 /// @brief Determine if it is valid to fold a cast of a cast
75 unsigned opc, ///< opcode of the second cast constant expression
76 ConstantExpr *Op, ///< the first cast constant expression
77 const Type *DstTy ///< desintation type of the first cast
79 assert(Op && Op->isCast() && "Can't fold cast of cast without a cast!");
80 assert(DstTy && DstTy->isFirstClassType() && "Invalid cast destination type");
81 assert(CastInst::isCast(opc) && "Invalid cast opcode");
83 // The the types and opcodes for the two Cast constant expressions
84 const Type *SrcTy = Op->getOperand(0)->getType();
85 const Type *MidTy = Op->getType();
86 Instruction::CastOps firstOp = Instruction::CastOps(Op->getOpcode());
87 Instruction::CastOps secondOp = Instruction::CastOps(opc);
89 // Let CastInst::isEliminableCastPair do the heavy lifting.
90 return CastInst::isEliminableCastPair(firstOp, secondOp, SrcTy, MidTy, DstTy,
91 Type::getInt64Ty(DstTy->getContext()));
94 static Constant *FoldBitCast(LLVMContext &Context,
95 Constant *V, const Type *DestTy) {
96 const Type *SrcTy = V->getType();
98 return V; // no-op cast
100 // Check to see if we are casting a pointer to an aggregate to a pointer to
101 // the first element. If so, return the appropriate GEP instruction.
102 if (const PointerType *PTy = dyn_cast<PointerType>(V->getType()))
103 if (const PointerType *DPTy = dyn_cast<PointerType>(DestTy))
104 if (PTy->getAddressSpace() == DPTy->getAddressSpace()) {
105 SmallVector<Value*, 8> IdxList;
106 Value *Zero = Constant::getNullValue(Type::getInt32Ty(Context));
107 IdxList.push_back(Zero);
108 const Type *ElTy = PTy->getElementType();
109 while (ElTy != DPTy->getElementType()) {
110 if (const StructType *STy = dyn_cast<StructType>(ElTy)) {
111 if (STy->getNumElements() == 0) break;
112 ElTy = STy->getElementType(0);
113 IdxList.push_back(Zero);
114 } else if (const SequentialType *STy =
115 dyn_cast<SequentialType>(ElTy)) {
116 if (isa<PointerType>(ElTy)) break; // Can't index into pointers!
117 ElTy = STy->getElementType();
118 IdxList.push_back(Zero);
124 if (ElTy == DPTy->getElementType())
125 // This GEP is inbounds because all indices are zero.
126 return ConstantExpr::getInBoundsGetElementPtr(V, &IdxList[0],
130 // Handle casts from one vector constant to another. We know that the src
131 // and dest type have the same size (otherwise its an illegal cast).
132 if (const VectorType *DestPTy = dyn_cast<VectorType>(DestTy)) {
133 if (const VectorType *SrcTy = dyn_cast<VectorType>(V->getType())) {
134 assert(DestPTy->getBitWidth() == SrcTy->getBitWidth() &&
135 "Not cast between same sized vectors!");
137 // First, check for null. Undef is already handled.
138 if (isa<ConstantAggregateZero>(V))
139 return Constant::getNullValue(DestTy);
141 if (ConstantVector *CV = dyn_cast<ConstantVector>(V))
142 return BitCastConstantVector(Context, CV, DestPTy);
145 // Canonicalize scalar-to-vector bitcasts into vector-to-vector bitcasts
146 // This allows for other simplifications (although some of them
147 // can only be handled by Analysis/ConstantFolding.cpp).
148 if (isa<ConstantInt>(V) || isa<ConstantFP>(V))
149 return ConstantExpr::getBitCast(
150 ConstantVector::get(&V, 1), DestPTy);
153 // Finally, implement bitcast folding now. The code below doesn't handle
155 if (isa<ConstantPointerNull>(V)) // ptr->ptr cast.
156 return ConstantPointerNull::get(cast<PointerType>(DestTy));
158 // Handle integral constant input.
159 if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
160 if (DestTy->isInteger())
161 // Integral -> Integral. This is a no-op because the bit widths must
162 // be the same. Consequently, we just fold to V.
165 if (DestTy->isFloatingPoint())
166 return ConstantFP::get(Context, APFloat(CI->getValue(),
167 DestTy != Type::getPPC_FP128Ty(Context)));
169 // Otherwise, can't fold this (vector?)
173 // Handle ConstantFP input.
174 if (ConstantFP *FP = dyn_cast<ConstantFP>(V))
176 return ConstantInt::get(Context, FP->getValueAPF().bitcastToAPInt());
182 /// ExtractConstantBytes - V is an integer constant which only has a subset of
183 /// its bytes used. The bytes used are indicated by ByteStart (which is the
184 /// first byte used, counting from the least significant byte) and ByteSize,
185 /// which is the number of bytes used.
187 /// This function analyzes the specified constant to see if the specified byte
188 /// range can be returned as a simplified constant. If so, the constant is
189 /// returned, otherwise null is returned.
191 static Constant *ExtractConstantBytes(Constant *C, unsigned ByteStart,
193 assert(isa<IntegerType>(C->getType()) &&
194 (cast<IntegerType>(C->getType())->getBitWidth() & 7) == 0 &&
195 "Non-byte sized integer input");
196 unsigned CSize = cast<IntegerType>(C->getType())->getBitWidth()/8;
197 assert(ByteSize && "Must be accessing some piece");
198 assert(ByteStart+ByteSize <= CSize && "Extracting invalid piece from input");
199 assert(ByteSize != CSize && "Should not extract everything");
201 // Constant Integers are simple.
202 if (ConstantInt *CI = dyn_cast<ConstantInt>(C)) {
203 APInt V = CI->getValue();
205 V = V.lshr(ByteStart*8);
207 return ConstantInt::get(CI->getContext(), V);
210 // In the input is a constant expr, we might be able to recursively simplify.
211 // If not, we definitely can't do anything.
212 ConstantExpr *CE = dyn_cast<ConstantExpr>(C);
213 if (CE == 0) return 0;
215 switch (CE->getOpcode()) {
217 case Instruction::Or: {
218 Constant *RHS = ExtractConstantBytes(CE->getOperand(1), ByteStart,ByteSize);
223 if (ConstantInt *RHSC = dyn_cast<ConstantInt>(RHS))
224 if (RHSC->isAllOnesValue())
227 Constant *LHS = ExtractConstantBytes(CE->getOperand(0), ByteStart,ByteSize);
230 return ConstantExpr::getOr(LHS, RHS);
232 case Instruction::And: {
233 Constant *RHS = ExtractConstantBytes(CE->getOperand(1), ByteStart,ByteSize);
238 if (RHS->isNullValue())
241 Constant *LHS = ExtractConstantBytes(CE->getOperand(0), ByteStart,ByteSize);
244 return ConstantExpr::getAnd(LHS, RHS);
246 case Instruction::LShr: {
247 ConstantInt *Amt = dyn_cast<ConstantInt>(CE->getOperand(1));
250 unsigned ShAmt = Amt->getZExtValue();
251 // Cannot analyze non-byte shifts.
252 if ((ShAmt & 7) != 0)
256 // If the extract is known to be all zeros, return zero.
257 if (ByteStart >= CSize-ShAmt)
258 return Constant::getNullValue(IntegerType::get(CE->getContext(),
260 // If the extract is known to be fully in the input, extract it.
261 if (ByteStart+ByteSize+ShAmt <= CSize)
262 return ExtractConstantBytes(CE->getOperand(0), ByteStart+ShAmt, ByteSize);
264 // TODO: Handle the 'partially zero' case.
268 case Instruction::Shl: {
269 ConstantInt *Amt = dyn_cast<ConstantInt>(CE->getOperand(1));
272 unsigned ShAmt = Amt->getZExtValue();
273 // Cannot analyze non-byte shifts.
274 if ((ShAmt & 7) != 0)
278 // If the extract is known to be all zeros, return zero.
279 if (ByteStart+ByteSize <= ShAmt)
280 return Constant::getNullValue(IntegerType::get(CE->getContext(),
282 // If the extract is known to be fully in the input, extract it.
283 if (ByteStart >= ShAmt)
284 return ExtractConstantBytes(CE->getOperand(0), ByteStart-ShAmt, ByteSize);
286 // TODO: Handle the 'partially zero' case.
290 case Instruction::ZExt: {
291 unsigned SrcBitSize =
292 cast<IntegerType>(CE->getOperand(0)->getType())->getBitWidth();
294 // If extracting something that is completely zero, return 0.
295 if (ByteStart*8 >= SrcBitSize)
296 return Constant::getNullValue(IntegerType::get(CE->getContext(),
299 // If exactly extracting the input, return it.
300 if (ByteStart == 0 && ByteSize*8 == SrcBitSize)
301 return CE->getOperand(0);
303 // If extracting something completely in the input, if if the input is a
304 // multiple of 8 bits, recurse.
305 if ((SrcBitSize&7) == 0 && (ByteStart+ByteSize)*8 <= SrcBitSize)
306 return ExtractConstantBytes(CE->getOperand(0), ByteStart, ByteSize);
308 // Otherwise, if extracting a subset of the input, which is not multiple of
309 // 8 bits, do a shift and trunc to get the bits.
310 if ((ByteStart+ByteSize)*8 < SrcBitSize) {
311 assert((SrcBitSize&7) && "Shouldn't get byte sized case here");
312 Constant *Res = CE->getOperand(0);
314 Res = ConstantExpr::getLShr(Res,
315 ConstantInt::get(Res->getType(), ByteStart*8));
316 return ConstantExpr::getTrunc(Res, IntegerType::get(C->getContext(),
320 // TODO: Handle the 'partially zero' case.
327 Constant *llvm::ConstantFoldCastInstruction(LLVMContext &Context,
328 unsigned opc, Constant *V,
329 const Type *DestTy) {
330 if (isa<UndefValue>(V)) {
331 // zext(undef) = 0, because the top bits will be zero.
332 // sext(undef) = 0, because the top bits will all be the same.
333 // [us]itofp(undef) = 0, because the result value is bounded.
334 if (opc == Instruction::ZExt || opc == Instruction::SExt ||
335 opc == Instruction::UIToFP || opc == Instruction::SIToFP)
336 return Constant::getNullValue(DestTy);
337 return UndefValue::get(DestTy);
339 // No compile-time operations on this type yet.
340 if (V->getType()->isPPC_FP128Ty() || DestTy->isPPC_FP128Ty())
343 // If the cast operand is a constant expression, there's a few things we can
344 // do to try to simplify it.
345 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V)) {
347 // Try hard to fold cast of cast because they are often eliminable.
348 if (unsigned newOpc = foldConstantCastPair(opc, CE, DestTy))
349 return ConstantExpr::getCast(newOpc, CE->getOperand(0), DestTy);
350 } else if (CE->getOpcode() == Instruction::GetElementPtr) {
351 // If all of the indexes in the GEP are null values, there is no pointer
352 // adjustment going on. We might as well cast the source pointer.
353 bool isAllNull = true;
354 for (unsigned i = 1, e = CE->getNumOperands(); i != e; ++i)
355 if (!CE->getOperand(i)->isNullValue()) {
360 // This is casting one pointer type to another, always BitCast
361 return ConstantExpr::getPointerCast(CE->getOperand(0), DestTy);
365 // If the cast operand is a constant vector, perform the cast by
366 // operating on each element. In the cast of bitcasts, the element
367 // count may be mismatched; don't attempt to handle that here.
368 if (ConstantVector *CV = dyn_cast<ConstantVector>(V))
369 if (isa<VectorType>(DestTy) &&
370 cast<VectorType>(DestTy)->getNumElements() ==
371 CV->getType()->getNumElements()) {
372 std::vector<Constant*> res;
373 const VectorType *DestVecTy = cast<VectorType>(DestTy);
374 const Type *DstEltTy = DestVecTy->getElementType();
375 for (unsigned i = 0, e = CV->getType()->getNumElements(); i != e; ++i)
376 res.push_back(ConstantExpr::getCast(opc,
377 CV->getOperand(i), DstEltTy));
378 return ConstantVector::get(DestVecTy, res);
381 // We actually have to do a cast now. Perform the cast according to the
385 llvm_unreachable("Failed to cast constant expression");
386 case Instruction::FPTrunc:
387 case Instruction::FPExt:
388 if (ConstantFP *FPC = dyn_cast<ConstantFP>(V)) {
390 APFloat Val = FPC->getValueAPF();
391 Val.convert(DestTy->isFloatTy() ? APFloat::IEEEsingle :
392 DestTy->isDoubleTy() ? APFloat::IEEEdouble :
393 DestTy->isX86_FP80Ty() ? APFloat::x87DoubleExtended :
394 DestTy->isFP128Ty() ? APFloat::IEEEquad :
396 APFloat::rmNearestTiesToEven, &ignored);
397 return ConstantFP::get(Context, Val);
399 return 0; // Can't fold.
400 case Instruction::FPToUI:
401 case Instruction::FPToSI:
402 if (ConstantFP *FPC = dyn_cast<ConstantFP>(V)) {
403 const APFloat &V = FPC->getValueAPF();
406 uint32_t DestBitWidth = cast<IntegerType>(DestTy)->getBitWidth();
407 (void) V.convertToInteger(x, DestBitWidth, opc==Instruction::FPToSI,
408 APFloat::rmTowardZero, &ignored);
409 APInt Val(DestBitWidth, 2, x);
410 return ConstantInt::get(Context, Val);
412 return 0; // Can't fold.
413 case Instruction::IntToPtr: //always treated as unsigned
414 if (V->isNullValue()) // Is it an integral null value?
415 return ConstantPointerNull::get(cast<PointerType>(DestTy));
416 return 0; // Other pointer types cannot be casted
417 case Instruction::PtrToInt: // always treated as unsigned
418 if (V->isNullValue()) // is it a null pointer value?
419 return ConstantInt::get(DestTy, 0);
420 return 0; // Other pointer types cannot be casted
421 case Instruction::UIToFP:
422 case Instruction::SIToFP:
423 if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
424 APInt api = CI->getValue();
425 const uint64_t zero[] = {0, 0};
426 APFloat apf = APFloat(APInt(DestTy->getPrimitiveSizeInBits(),
428 (void)apf.convertFromAPInt(api,
429 opc==Instruction::SIToFP,
430 APFloat::rmNearestTiesToEven);
431 return ConstantFP::get(Context, apf);
434 case Instruction::ZExt:
435 if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
436 uint32_t BitWidth = cast<IntegerType>(DestTy)->getBitWidth();
437 APInt Result(CI->getValue());
438 Result.zext(BitWidth);
439 return ConstantInt::get(Context, Result);
442 case Instruction::SExt:
443 if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
444 uint32_t BitWidth = cast<IntegerType>(DestTy)->getBitWidth();
445 APInt Result(CI->getValue());
446 Result.sext(BitWidth);
447 return ConstantInt::get(Context, Result);
450 case Instruction::Trunc: {
451 uint32_t DestBitWidth = cast<IntegerType>(DestTy)->getBitWidth();
452 if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
453 APInt Result(CI->getValue());
454 Result.trunc(DestBitWidth);
455 return ConstantInt::get(Context, Result);
458 // The input must be a constantexpr. See if we can simplify this based on
459 // the bytes we are demanding. Only do this if the source and dest are an
460 // even multiple of a byte.
461 if ((DestBitWidth & 7) == 0 &&
462 (cast<IntegerType>(V->getType())->getBitWidth() & 7) == 0)
463 if (Constant *Res = ExtractConstantBytes(V, 0, DestBitWidth / 8))
468 case Instruction::BitCast:
469 return FoldBitCast(Context, V, DestTy);
473 Constant *llvm::ConstantFoldSelectInstruction(LLVMContext&,
475 Constant *V1, Constant *V2) {
476 if (ConstantInt *CB = dyn_cast<ConstantInt>(Cond))
477 return CB->getZExtValue() ? V1 : V2;
479 if (isa<UndefValue>(V1)) return V2;
480 if (isa<UndefValue>(V2)) return V1;
481 if (isa<UndefValue>(Cond)) return V1;
482 if (V1 == V2) return V1;
486 Constant *llvm::ConstantFoldExtractElementInstruction(LLVMContext &Context,
489 if (isa<UndefValue>(Val)) // ee(undef, x) -> undef
490 return UndefValue::get(cast<VectorType>(Val->getType())->getElementType());
491 if (Val->isNullValue()) // ee(zero, x) -> zero
492 return Constant::getNullValue(
493 cast<VectorType>(Val->getType())->getElementType());
495 if (ConstantVector *CVal = dyn_cast<ConstantVector>(Val)) {
496 if (ConstantInt *CIdx = dyn_cast<ConstantInt>(Idx)) {
497 return CVal->getOperand(CIdx->getZExtValue());
498 } else if (isa<UndefValue>(Idx)) {
499 // ee({w,x,y,z}, undef) -> w (an arbitrary value).
500 return CVal->getOperand(0);
506 Constant *llvm::ConstantFoldInsertElementInstruction(LLVMContext &Context,
510 ConstantInt *CIdx = dyn_cast<ConstantInt>(Idx);
512 APInt idxVal = CIdx->getValue();
513 if (isa<UndefValue>(Val)) {
514 // Insertion of scalar constant into vector undef
515 // Optimize away insertion of undef
516 if (isa<UndefValue>(Elt))
518 // Otherwise break the aggregate undef into multiple undefs and do
521 cast<VectorType>(Val->getType())->getNumElements();
522 std::vector<Constant*> Ops;
524 for (unsigned i = 0; i < numOps; ++i) {
526 (idxVal == i) ? Elt : UndefValue::get(Elt->getType());
529 return ConstantVector::get(Ops);
531 if (isa<ConstantAggregateZero>(Val)) {
532 // Insertion of scalar constant into vector aggregate zero
533 // Optimize away insertion of zero
534 if (Elt->isNullValue())
536 // Otherwise break the aggregate zero into multiple zeros and do
539 cast<VectorType>(Val->getType())->getNumElements();
540 std::vector<Constant*> Ops;
542 for (unsigned i = 0; i < numOps; ++i) {
544 (idxVal == i) ? Elt : Constant::getNullValue(Elt->getType());
547 return ConstantVector::get(Ops);
549 if (ConstantVector *CVal = dyn_cast<ConstantVector>(Val)) {
550 // Insertion of scalar constant into vector constant
551 std::vector<Constant*> Ops;
552 Ops.reserve(CVal->getNumOperands());
553 for (unsigned i = 0; i < CVal->getNumOperands(); ++i) {
555 (idxVal == i) ? Elt : cast<Constant>(CVal->getOperand(i));
558 return ConstantVector::get(Ops);
564 /// GetVectorElement - If C is a ConstantVector, ConstantAggregateZero or Undef
565 /// return the specified element value. Otherwise return null.
566 static Constant *GetVectorElement(LLVMContext &Context, Constant *C,
568 if (ConstantVector *CV = dyn_cast<ConstantVector>(C))
569 return CV->getOperand(EltNo);
571 const Type *EltTy = cast<VectorType>(C->getType())->getElementType();
572 if (isa<ConstantAggregateZero>(C))
573 return Constant::getNullValue(EltTy);
574 if (isa<UndefValue>(C))
575 return UndefValue::get(EltTy);
579 Constant *llvm::ConstantFoldShuffleVectorInstruction(LLVMContext &Context,
583 // Undefined shuffle mask -> undefined value.
584 if (isa<UndefValue>(Mask)) return UndefValue::get(V1->getType());
586 unsigned MaskNumElts = cast<VectorType>(Mask->getType())->getNumElements();
587 unsigned SrcNumElts = cast<VectorType>(V1->getType())->getNumElements();
588 const Type *EltTy = cast<VectorType>(V1->getType())->getElementType();
590 // Loop over the shuffle mask, evaluating each element.
591 SmallVector<Constant*, 32> Result;
592 for (unsigned i = 0; i != MaskNumElts; ++i) {
593 Constant *InElt = GetVectorElement(Context, Mask, i);
594 if (InElt == 0) return 0;
596 if (isa<UndefValue>(InElt))
597 InElt = UndefValue::get(EltTy);
598 else if (ConstantInt *CI = dyn_cast<ConstantInt>(InElt)) {
599 unsigned Elt = CI->getZExtValue();
600 if (Elt >= SrcNumElts*2)
601 InElt = UndefValue::get(EltTy);
602 else if (Elt >= SrcNumElts)
603 InElt = GetVectorElement(Context, V2, Elt - SrcNumElts);
605 InElt = GetVectorElement(Context, V1, Elt);
606 if (InElt == 0) return 0;
611 Result.push_back(InElt);
614 return ConstantVector::get(&Result[0], Result.size());
617 Constant *llvm::ConstantFoldExtractValueInstruction(LLVMContext &Context,
619 const unsigned *Idxs,
621 // Base case: no indices, so return the entire value.
625 if (isa<UndefValue>(Agg)) // ev(undef, x) -> undef
626 return UndefValue::get(ExtractValueInst::getIndexedType(Agg->getType(),
630 if (isa<ConstantAggregateZero>(Agg)) // ev(0, x) -> 0
632 Constant::getNullValue(ExtractValueInst::getIndexedType(Agg->getType(),
636 // Otherwise recurse.
637 if (ConstantStruct *CS = dyn_cast<ConstantStruct>(Agg))
638 return ConstantFoldExtractValueInstruction(Context, CS->getOperand(*Idxs),
641 if (ConstantArray *CA = dyn_cast<ConstantArray>(Agg))
642 return ConstantFoldExtractValueInstruction(Context, CA->getOperand(*Idxs),
644 ConstantVector *CV = cast<ConstantVector>(Agg);
645 return ConstantFoldExtractValueInstruction(Context, CV->getOperand(*Idxs),
649 Constant *llvm::ConstantFoldInsertValueInstruction(LLVMContext &Context,
652 const unsigned *Idxs,
654 // Base case: no indices, so replace the entire value.
658 if (isa<UndefValue>(Agg)) {
659 // Insertion of constant into aggregate undef
660 // Optimize away insertion of undef.
661 if (isa<UndefValue>(Val))
664 // Otherwise break the aggregate undef into multiple undefs and do
666 const CompositeType *AggTy = cast<CompositeType>(Agg->getType());
668 if (const ArrayType *AR = dyn_cast<ArrayType>(AggTy))
669 numOps = AR->getNumElements();
671 numOps = cast<StructType>(AggTy)->getNumElements();
673 std::vector<Constant*> Ops(numOps);
674 for (unsigned i = 0; i < numOps; ++i) {
675 const Type *MemberTy = AggTy->getTypeAtIndex(i);
678 ConstantFoldInsertValueInstruction(Context, UndefValue::get(MemberTy),
679 Val, Idxs+1, NumIdx-1) :
680 UndefValue::get(MemberTy);
684 if (const StructType* ST = dyn_cast<StructType>(AggTy))
685 return ConstantStruct::get(Context, Ops, ST->isPacked());
686 return ConstantArray::get(cast<ArrayType>(AggTy), Ops);
689 if (isa<ConstantAggregateZero>(Agg)) {
690 // Insertion of constant into aggregate zero
691 // Optimize away insertion of zero.
692 if (Val->isNullValue())
695 // Otherwise break the aggregate zero into multiple zeros and do
697 const CompositeType *AggTy = cast<CompositeType>(Agg->getType());
699 if (const ArrayType *AR = dyn_cast<ArrayType>(AggTy))
700 numOps = AR->getNumElements();
702 numOps = cast<StructType>(AggTy)->getNumElements();
704 std::vector<Constant*> Ops(numOps);
705 for (unsigned i = 0; i < numOps; ++i) {
706 const Type *MemberTy = AggTy->getTypeAtIndex(i);
709 ConstantFoldInsertValueInstruction(Context,
710 Constant::getNullValue(MemberTy),
711 Val, Idxs+1, NumIdx-1) :
712 Constant::getNullValue(MemberTy);
716 if (const StructType* ST = dyn_cast<StructType>(AggTy))
717 return ConstantStruct::get(Context, Ops, ST->isPacked());
718 return ConstantArray::get(cast<ArrayType>(AggTy), Ops);
721 if (isa<ConstantStruct>(Agg) || isa<ConstantArray>(Agg)) {
722 // Insertion of constant into aggregate constant.
723 std::vector<Constant*> Ops(Agg->getNumOperands());
724 for (unsigned i = 0; i < Agg->getNumOperands(); ++i) {
725 Constant *Op = cast<Constant>(Agg->getOperand(i));
727 Op = ConstantFoldInsertValueInstruction(Context, Op,
728 Val, Idxs+1, NumIdx-1);
732 if (const StructType* ST = dyn_cast<StructType>(Agg->getType()))
733 return ConstantStruct::get(Context, Ops, ST->isPacked());
734 return ConstantArray::get(cast<ArrayType>(Agg->getType()), Ops);
741 Constant *llvm::ConstantFoldBinaryInstruction(LLVMContext &Context,
743 Constant *C1, Constant *C2) {
744 // No compile-time operations on this type yet.
745 if (C1->getType()->isPPC_FP128Ty())
748 // Handle UndefValue up front.
749 if (isa<UndefValue>(C1) || isa<UndefValue>(C2)) {
751 case Instruction::Xor:
752 if (isa<UndefValue>(C1) && isa<UndefValue>(C2))
753 // Handle undef ^ undef -> 0 special case. This is a common
755 return Constant::getNullValue(C1->getType());
757 case Instruction::Add:
758 case Instruction::Sub:
759 return UndefValue::get(C1->getType());
760 case Instruction::Mul:
761 case Instruction::And:
762 return Constant::getNullValue(C1->getType());
763 case Instruction::UDiv:
764 case Instruction::SDiv:
765 case Instruction::URem:
766 case Instruction::SRem:
767 if (!isa<UndefValue>(C2)) // undef / X -> 0
768 return Constant::getNullValue(C1->getType());
769 return C2; // X / undef -> undef
770 case Instruction::Or: // X | undef -> -1
771 if (const VectorType *PTy = dyn_cast<VectorType>(C1->getType()))
772 return Constant::getAllOnesValue(PTy);
773 return Constant::getAllOnesValue(C1->getType());
774 case Instruction::LShr:
775 if (isa<UndefValue>(C2) && isa<UndefValue>(C1))
776 return C1; // undef lshr undef -> undef
777 return Constant::getNullValue(C1->getType()); // X lshr undef -> 0
779 case Instruction::AShr:
780 if (!isa<UndefValue>(C2))
781 return C1; // undef ashr X --> undef
782 else if (isa<UndefValue>(C1))
783 return C1; // undef ashr undef -> undef
785 return C1; // X ashr undef --> X
786 case Instruction::Shl:
787 // undef << X -> 0 or X << undef -> 0
788 return Constant::getNullValue(C1->getType());
792 // Handle simplifications when the RHS is a constant int.
793 if (ConstantInt *CI2 = dyn_cast<ConstantInt>(C2)) {
795 case Instruction::Add:
796 if (CI2->equalsInt(0)) return C1; // X + 0 == X
798 case Instruction::Sub:
799 if (CI2->equalsInt(0)) return C1; // X - 0 == X
801 case Instruction::Mul:
802 if (CI2->equalsInt(0)) return C2; // X * 0 == 0
803 if (CI2->equalsInt(1))
804 return C1; // X * 1 == X
806 case Instruction::UDiv:
807 case Instruction::SDiv:
808 if (CI2->equalsInt(1))
809 return C1; // X / 1 == X
810 if (CI2->equalsInt(0))
811 return UndefValue::get(CI2->getType()); // X / 0 == undef
813 case Instruction::URem:
814 case Instruction::SRem:
815 if (CI2->equalsInt(1))
816 return Constant::getNullValue(CI2->getType()); // X % 1 == 0
817 if (CI2->equalsInt(0))
818 return UndefValue::get(CI2->getType()); // X % 0 == undef
820 case Instruction::And:
821 if (CI2->isZero()) return C2; // X & 0 == 0
822 if (CI2->isAllOnesValue())
823 return C1; // X & -1 == X
825 if (ConstantExpr *CE1 = dyn_cast<ConstantExpr>(C1)) {
826 // (zext i32 to i64) & 4294967295 -> (zext i32 to i64)
827 if (CE1->getOpcode() == Instruction::ZExt) {
828 unsigned DstWidth = CI2->getType()->getBitWidth();
830 CE1->getOperand(0)->getType()->getPrimitiveSizeInBits();
831 APInt PossiblySetBits(APInt::getLowBitsSet(DstWidth, SrcWidth));
832 if ((PossiblySetBits & CI2->getValue()) == PossiblySetBits)
836 // If and'ing the address of a global with a constant, fold it.
837 if (CE1->getOpcode() == Instruction::PtrToInt &&
838 isa<GlobalValue>(CE1->getOperand(0))) {
839 GlobalValue *GV = cast<GlobalValue>(CE1->getOperand(0));
841 // Functions are at least 4-byte aligned.
842 unsigned GVAlign = GV->getAlignment();
843 if (isa<Function>(GV))
844 GVAlign = std::max(GVAlign, 4U);
847 unsigned DstWidth = CI2->getType()->getBitWidth();
848 unsigned SrcWidth = std::min(DstWidth, Log2_32(GVAlign));
849 APInt BitsNotSet(APInt::getLowBitsSet(DstWidth, SrcWidth));
851 // If checking bits we know are clear, return zero.
852 if ((CI2->getValue() & BitsNotSet) == CI2->getValue())
853 return Constant::getNullValue(CI2->getType());
858 case Instruction::Or:
859 if (CI2->equalsInt(0)) return C1; // X | 0 == X
860 if (CI2->isAllOnesValue())
861 return C2; // X | -1 == -1
863 case Instruction::Xor:
864 if (CI2->equalsInt(0)) return C1; // X ^ 0 == X
866 if (ConstantExpr *CE1 = dyn_cast<ConstantExpr>(C1)) {
867 switch (CE1->getOpcode()) {
869 case Instruction::ICmp:
870 case Instruction::FCmp:
871 // cmp pred ^ true -> cmp !pred
872 assert(CI2->equalsInt(1));
873 CmpInst::Predicate pred = (CmpInst::Predicate)CE1->getPredicate();
874 pred = CmpInst::getInversePredicate(pred);
875 return ConstantExpr::getCompare(pred, CE1->getOperand(0),
880 case Instruction::AShr:
881 // ashr (zext C to Ty), C2 -> lshr (zext C, CSA), C2
882 if (ConstantExpr *CE1 = dyn_cast<ConstantExpr>(C1))
883 if (CE1->getOpcode() == Instruction::ZExt) // Top bits known zero.
884 return ConstantExpr::getLShr(C1, C2);
889 // At this point we know neither constant is an UndefValue.
890 if (ConstantInt *CI1 = dyn_cast<ConstantInt>(C1)) {
891 if (ConstantInt *CI2 = dyn_cast<ConstantInt>(C2)) {
892 using namespace APIntOps;
893 const APInt &C1V = CI1->getValue();
894 const APInt &C2V = CI2->getValue();
898 case Instruction::Add:
899 return ConstantInt::get(Context, C1V + C2V);
900 case Instruction::Sub:
901 return ConstantInt::get(Context, C1V - C2V);
902 case Instruction::Mul:
903 return ConstantInt::get(Context, C1V * C2V);
904 case Instruction::UDiv:
905 assert(!CI2->isNullValue() && "Div by zero handled above");
906 return ConstantInt::get(Context, C1V.udiv(C2V));
907 case Instruction::SDiv:
908 assert(!CI2->isNullValue() && "Div by zero handled above");
909 if (C2V.isAllOnesValue() && C1V.isMinSignedValue())
910 return UndefValue::get(CI1->getType()); // MIN_INT / -1 -> undef
911 return ConstantInt::get(Context, C1V.sdiv(C2V));
912 case Instruction::URem:
913 assert(!CI2->isNullValue() && "Div by zero handled above");
914 return ConstantInt::get(Context, C1V.urem(C2V));
915 case Instruction::SRem:
916 assert(!CI2->isNullValue() && "Div by zero handled above");
917 if (C2V.isAllOnesValue() && C1V.isMinSignedValue())
918 return UndefValue::get(CI1->getType()); // MIN_INT % -1 -> undef
919 return ConstantInt::get(Context, C1V.srem(C2V));
920 case Instruction::And:
921 return ConstantInt::get(Context, C1V & C2V);
922 case Instruction::Or:
923 return ConstantInt::get(Context, C1V | C2V);
924 case Instruction::Xor:
925 return ConstantInt::get(Context, C1V ^ C2V);
926 case Instruction::Shl: {
927 uint32_t shiftAmt = C2V.getZExtValue();
928 if (shiftAmt < C1V.getBitWidth())
929 return ConstantInt::get(Context, C1V.shl(shiftAmt));
931 return UndefValue::get(C1->getType()); // too big shift is undef
933 case Instruction::LShr: {
934 uint32_t shiftAmt = C2V.getZExtValue();
935 if (shiftAmt < C1V.getBitWidth())
936 return ConstantInt::get(Context, C1V.lshr(shiftAmt));
938 return UndefValue::get(C1->getType()); // too big shift is undef
940 case Instruction::AShr: {
941 uint32_t shiftAmt = C2V.getZExtValue();
942 if (shiftAmt < C1V.getBitWidth())
943 return ConstantInt::get(Context, C1V.ashr(shiftAmt));
945 return UndefValue::get(C1->getType()); // too big shift is undef
951 case Instruction::SDiv:
952 case Instruction::UDiv:
953 case Instruction::URem:
954 case Instruction::SRem:
955 case Instruction::LShr:
956 case Instruction::AShr:
957 case Instruction::Shl:
958 if (CI1->equalsInt(0)) return C1;
963 } else if (ConstantFP *CFP1 = dyn_cast<ConstantFP>(C1)) {
964 if (ConstantFP *CFP2 = dyn_cast<ConstantFP>(C2)) {
965 APFloat C1V = CFP1->getValueAPF();
966 APFloat C2V = CFP2->getValueAPF();
967 APFloat C3V = C1V; // copy for modification
971 case Instruction::FAdd:
972 (void)C3V.add(C2V, APFloat::rmNearestTiesToEven);
973 return ConstantFP::get(Context, C3V);
974 case Instruction::FSub:
975 (void)C3V.subtract(C2V, APFloat::rmNearestTiesToEven);
976 return ConstantFP::get(Context, C3V);
977 case Instruction::FMul:
978 (void)C3V.multiply(C2V, APFloat::rmNearestTiesToEven);
979 return ConstantFP::get(Context, C3V);
980 case Instruction::FDiv:
981 (void)C3V.divide(C2V, APFloat::rmNearestTiesToEven);
982 return ConstantFP::get(Context, C3V);
983 case Instruction::FRem:
984 (void)C3V.mod(C2V, APFloat::rmNearestTiesToEven);
985 return ConstantFP::get(Context, C3V);
988 } else if (const VectorType *VTy = dyn_cast<VectorType>(C1->getType())) {
989 ConstantVector *CP1 = dyn_cast<ConstantVector>(C1);
990 ConstantVector *CP2 = dyn_cast<ConstantVector>(C2);
991 if ((CP1 != NULL || isa<ConstantAggregateZero>(C1)) &&
992 (CP2 != NULL || isa<ConstantAggregateZero>(C2))) {
993 std::vector<Constant*> Res;
994 const Type* EltTy = VTy->getElementType();
1000 case Instruction::Add:
1001 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
1002 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
1003 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
1004 Res.push_back(ConstantExpr::getAdd(C1, C2));
1006 return ConstantVector::get(Res);
1007 case Instruction::FAdd:
1008 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
1009 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
1010 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
1011 Res.push_back(ConstantExpr::getFAdd(C1, C2));
1013 return ConstantVector::get(Res);
1014 case Instruction::Sub:
1015 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
1016 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
1017 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
1018 Res.push_back(ConstantExpr::getSub(C1, C2));
1020 return ConstantVector::get(Res);
1021 case Instruction::FSub:
1022 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
1023 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
1024 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
1025 Res.push_back(ConstantExpr::getFSub(C1, C2));
1027 return ConstantVector::get(Res);
1028 case Instruction::Mul:
1029 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
1030 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
1031 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
1032 Res.push_back(ConstantExpr::getMul(C1, C2));
1034 return ConstantVector::get(Res);
1035 case Instruction::FMul:
1036 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
1037 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
1038 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
1039 Res.push_back(ConstantExpr::getFMul(C1, C2));
1041 return ConstantVector::get(Res);
1042 case Instruction::UDiv:
1043 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
1044 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
1045 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
1046 Res.push_back(ConstantExpr::getUDiv(C1, C2));
1048 return ConstantVector::get(Res);
1049 case Instruction::SDiv:
1050 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
1051 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
1052 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
1053 Res.push_back(ConstantExpr::getSDiv(C1, C2));
1055 return ConstantVector::get(Res);
1056 case Instruction::FDiv:
1057 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
1058 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
1059 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
1060 Res.push_back(ConstantExpr::getFDiv(C1, C2));
1062 return ConstantVector::get(Res);
1063 case Instruction::URem:
1064 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
1065 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
1066 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
1067 Res.push_back(ConstantExpr::getURem(C1, C2));
1069 return ConstantVector::get(Res);
1070 case Instruction::SRem:
1071 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
1072 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
1073 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
1074 Res.push_back(ConstantExpr::getSRem(C1, C2));
1076 return ConstantVector::get(Res);
1077 case Instruction::FRem:
1078 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
1079 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
1080 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
1081 Res.push_back(ConstantExpr::getFRem(C1, C2));
1083 return ConstantVector::get(Res);
1084 case Instruction::And:
1085 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
1086 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
1087 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
1088 Res.push_back(ConstantExpr::getAnd(C1, C2));
1090 return ConstantVector::get(Res);
1091 case Instruction::Or:
1092 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
1093 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
1094 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
1095 Res.push_back(ConstantExpr::getOr(C1, C2));
1097 return ConstantVector::get(Res);
1098 case Instruction::Xor:
1099 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
1100 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
1101 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
1102 Res.push_back(ConstantExpr::getXor(C1, C2));
1104 return ConstantVector::get(Res);
1105 case Instruction::LShr:
1106 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
1107 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
1108 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
1109 Res.push_back(ConstantExpr::getLShr(C1, C2));
1111 return ConstantVector::get(Res);
1112 case Instruction::AShr:
1113 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
1114 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
1115 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
1116 Res.push_back(ConstantExpr::getAShr(C1, C2));
1118 return ConstantVector::get(Res);
1119 case Instruction::Shl:
1120 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
1121 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
1122 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
1123 Res.push_back(ConstantExpr::getShl(C1, C2));
1125 return ConstantVector::get(Res);
1130 if (isa<ConstantExpr>(C1)) {
1131 // There are many possible foldings we could do here. We should probably
1132 // at least fold add of a pointer with an integer into the appropriate
1133 // getelementptr. This will improve alias analysis a bit.
1134 } else if (isa<ConstantExpr>(C2)) {
1135 // If C2 is a constant expr and C1 isn't, flop them around and fold the
1136 // other way if possible.
1138 case Instruction::Add:
1139 case Instruction::FAdd:
1140 case Instruction::Mul:
1141 case Instruction::FMul:
1142 case Instruction::And:
1143 case Instruction::Or:
1144 case Instruction::Xor:
1145 // No change of opcode required.
1146 return ConstantFoldBinaryInstruction(Context, Opcode, C2, C1);
1148 case Instruction::Shl:
1149 case Instruction::LShr:
1150 case Instruction::AShr:
1151 case Instruction::Sub:
1152 case Instruction::FSub:
1153 case Instruction::SDiv:
1154 case Instruction::UDiv:
1155 case Instruction::FDiv:
1156 case Instruction::URem:
1157 case Instruction::SRem:
1158 case Instruction::FRem:
1159 default: // These instructions cannot be flopped around.
1164 // i1 can be simplified in many cases.
1165 if (C1->getType() == Type::getInt1Ty(Context)) {
1167 case Instruction::Add:
1168 case Instruction::Sub:
1169 return ConstantExpr::getXor(C1, C2);
1170 case Instruction::Mul:
1171 return ConstantExpr::getAnd(C1, C2);
1172 case Instruction::Shl:
1173 case Instruction::LShr:
1174 case Instruction::AShr:
1175 // We can assume that C2 == 0. If it were one the result would be
1176 // undefined because the shift value is as large as the bitwidth.
1178 case Instruction::SDiv:
1179 case Instruction::UDiv:
1180 // We can assume that C2 == 1. If it were zero the result would be
1181 // undefined through division by zero.
1183 case Instruction::URem:
1184 case Instruction::SRem:
1185 // We can assume that C2 == 1. If it were zero the result would be
1186 // undefined through division by zero.
1187 return ConstantInt::getFalse(Context);
1193 // We don't know how to fold this.
1197 /// isZeroSizedType - This type is zero sized if its an array or structure of
1198 /// zero sized types. The only leaf zero sized type is an empty structure.
1199 static bool isMaybeZeroSizedType(const Type *Ty) {
1200 if (isa<OpaqueType>(Ty)) return true; // Can't say.
1201 if (const StructType *STy = dyn_cast<StructType>(Ty)) {
1203 // If all of elements have zero size, this does too.
1204 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i)
1205 if (!isMaybeZeroSizedType(STy->getElementType(i))) return false;
1208 } else if (const ArrayType *ATy = dyn_cast<ArrayType>(Ty)) {
1209 return isMaybeZeroSizedType(ATy->getElementType());
1214 /// IdxCompare - Compare the two constants as though they were getelementptr
1215 /// indices. This allows coersion of the types to be the same thing.
1217 /// If the two constants are the "same" (after coersion), return 0. If the
1218 /// first is less than the second, return -1, if the second is less than the
1219 /// first, return 1. If the constants are not integral, return -2.
1221 static int IdxCompare(LLVMContext &Context, Constant *C1, Constant *C2,
1223 if (C1 == C2) return 0;
1225 // Ok, we found a different index. If they are not ConstantInt, we can't do
1226 // anything with them.
1227 if (!isa<ConstantInt>(C1) || !isa<ConstantInt>(C2))
1228 return -2; // don't know!
1230 // Ok, we have two differing integer indices. Sign extend them to be the same
1231 // type. Long is always big enough, so we use it.
1232 if (C1->getType() != Type::getInt64Ty(Context))
1233 C1 = ConstantExpr::getSExt(C1, Type::getInt64Ty(Context));
1235 if (C2->getType() != Type::getInt64Ty(Context))
1236 C2 = ConstantExpr::getSExt(C2, Type::getInt64Ty(Context));
1238 if (C1 == C2) return 0; // They are equal
1240 // If the type being indexed over is really just a zero sized type, there is
1241 // no pointer difference being made here.
1242 if (isMaybeZeroSizedType(ElTy))
1243 return -2; // dunno.
1245 // If they are really different, now that they are the same type, then we
1246 // found a difference!
1247 if (cast<ConstantInt>(C1)->getSExtValue() <
1248 cast<ConstantInt>(C2)->getSExtValue())
1254 /// evaluateFCmpRelation - This function determines if there is anything we can
1255 /// decide about the two constants provided. This doesn't need to handle simple
1256 /// things like ConstantFP comparisons, but should instead handle ConstantExprs.
1257 /// If we can determine that the two constants have a particular relation to
1258 /// each other, we should return the corresponding FCmpInst predicate,
1259 /// otherwise return FCmpInst::BAD_FCMP_PREDICATE. This is used below in
1260 /// ConstantFoldCompareInstruction.
1262 /// To simplify this code we canonicalize the relation so that the first
1263 /// operand is always the most "complex" of the two. We consider ConstantFP
1264 /// to be the simplest, and ConstantExprs to be the most complex.
1265 static FCmpInst::Predicate evaluateFCmpRelation(LLVMContext &Context,
1266 Constant *V1, Constant *V2) {
1267 assert(V1->getType() == V2->getType() &&
1268 "Cannot compare values of different types!");
1270 // No compile-time operations on this type yet.
1271 if (V1->getType()->isPPC_FP128Ty())
1272 return FCmpInst::BAD_FCMP_PREDICATE;
1274 // Handle degenerate case quickly
1275 if (V1 == V2) return FCmpInst::FCMP_OEQ;
1277 if (!isa<ConstantExpr>(V1)) {
1278 if (!isa<ConstantExpr>(V2)) {
1279 // We distilled thisUse the standard constant folder for a few cases
1281 R = dyn_cast<ConstantInt>(
1282 ConstantExpr::getFCmp(FCmpInst::FCMP_OEQ, V1, V2));
1283 if (R && !R->isZero())
1284 return FCmpInst::FCMP_OEQ;
1285 R = dyn_cast<ConstantInt>(
1286 ConstantExpr::getFCmp(FCmpInst::FCMP_OLT, V1, V2));
1287 if (R && !R->isZero())
1288 return FCmpInst::FCMP_OLT;
1289 R = dyn_cast<ConstantInt>(
1290 ConstantExpr::getFCmp(FCmpInst::FCMP_OGT, V1, V2));
1291 if (R && !R->isZero())
1292 return FCmpInst::FCMP_OGT;
1294 // Nothing more we can do
1295 return FCmpInst::BAD_FCMP_PREDICATE;
1298 // If the first operand is simple and second is ConstantExpr, swap operands.
1299 FCmpInst::Predicate SwappedRelation = evaluateFCmpRelation(Context, V2, V1);
1300 if (SwappedRelation != FCmpInst::BAD_FCMP_PREDICATE)
1301 return FCmpInst::getSwappedPredicate(SwappedRelation);
1303 // Ok, the LHS is known to be a constantexpr. The RHS can be any of a
1304 // constantexpr or a simple constant.
1305 ConstantExpr *CE1 = cast<ConstantExpr>(V1);
1306 switch (CE1->getOpcode()) {
1307 case Instruction::FPTrunc:
1308 case Instruction::FPExt:
1309 case Instruction::UIToFP:
1310 case Instruction::SIToFP:
1311 // We might be able to do something with these but we don't right now.
1317 // There are MANY other foldings that we could perform here. They will
1318 // probably be added on demand, as they seem needed.
1319 return FCmpInst::BAD_FCMP_PREDICATE;
1322 /// evaluateICmpRelation - This function determines if there is anything we can
1323 /// decide about the two constants provided. This doesn't need to handle simple
1324 /// things like integer comparisons, but should instead handle ConstantExprs
1325 /// and GlobalValues. If we can determine that the two constants have a
1326 /// particular relation to each other, we should return the corresponding ICmp
1327 /// predicate, otherwise return ICmpInst::BAD_ICMP_PREDICATE.
1329 /// To simplify this code we canonicalize the relation so that the first
1330 /// operand is always the most "complex" of the two. We consider simple
1331 /// constants (like ConstantInt) to be the simplest, followed by
1332 /// GlobalValues, followed by ConstantExpr's (the most complex).
1334 static ICmpInst::Predicate evaluateICmpRelation(LLVMContext &Context,
1338 assert(V1->getType() == V2->getType() &&
1339 "Cannot compare different types of values!");
1340 if (V1 == V2) return ICmpInst::ICMP_EQ;
1342 if (!isa<ConstantExpr>(V1) && !isa<GlobalValue>(V1)) {
1343 if (!isa<GlobalValue>(V2) && !isa<ConstantExpr>(V2)) {
1344 // We distilled this down to a simple case, use the standard constant
1347 ICmpInst::Predicate pred = ICmpInst::ICMP_EQ;
1348 R = dyn_cast<ConstantInt>(ConstantExpr::getICmp(pred, V1, V2));
1349 if (R && !R->isZero())
1351 pred = isSigned ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT;
1352 R = dyn_cast<ConstantInt>(ConstantExpr::getICmp(pred, V1, V2));
1353 if (R && !R->isZero())
1355 pred = isSigned ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT;
1356 R = dyn_cast<ConstantInt>(ConstantExpr::getICmp(pred, V1, V2));
1357 if (R && !R->isZero())
1360 // If we couldn't figure it out, bail.
1361 return ICmpInst::BAD_ICMP_PREDICATE;
1364 // If the first operand is simple, swap operands.
1365 ICmpInst::Predicate SwappedRelation =
1366 evaluateICmpRelation(Context, V2, V1, isSigned);
1367 if (SwappedRelation != ICmpInst::BAD_ICMP_PREDICATE)
1368 return ICmpInst::getSwappedPredicate(SwappedRelation);
1370 } else if (const GlobalValue *CPR1 = dyn_cast<GlobalValue>(V1)) {
1371 if (isa<ConstantExpr>(V2)) { // Swap as necessary.
1372 ICmpInst::Predicate SwappedRelation =
1373 evaluateICmpRelation(Context, V2, V1, isSigned);
1374 if (SwappedRelation != ICmpInst::BAD_ICMP_PREDICATE)
1375 return ICmpInst::getSwappedPredicate(SwappedRelation);
1377 return ICmpInst::BAD_ICMP_PREDICATE;
1380 // Now we know that the RHS is a GlobalValue or simple constant,
1381 // which (since the types must match) means that it's a ConstantPointerNull.
1382 if (const GlobalValue *CPR2 = dyn_cast<GlobalValue>(V2)) {
1383 // Don't try to decide equality of aliases.
1384 if (!isa<GlobalAlias>(CPR1) && !isa<GlobalAlias>(CPR2))
1385 if (!CPR1->hasExternalWeakLinkage() || !CPR2->hasExternalWeakLinkage())
1386 return ICmpInst::ICMP_NE;
1388 assert(isa<ConstantPointerNull>(V2) && "Canonicalization guarantee!");
1389 // GlobalVals can never be null. Don't try to evaluate aliases.
1390 if (!CPR1->hasExternalWeakLinkage() && !isa<GlobalAlias>(CPR1))
1391 return ICmpInst::ICMP_NE;
1394 // Ok, the LHS is known to be a constantexpr. The RHS can be any of a
1395 // constantexpr, a CPR, or a simple constant.
1396 ConstantExpr *CE1 = cast<ConstantExpr>(V1);
1397 Constant *CE1Op0 = CE1->getOperand(0);
1399 switch (CE1->getOpcode()) {
1400 case Instruction::Trunc:
1401 case Instruction::FPTrunc:
1402 case Instruction::FPExt:
1403 case Instruction::FPToUI:
1404 case Instruction::FPToSI:
1405 break; // We can't evaluate floating point casts or truncations.
1407 case Instruction::UIToFP:
1408 case Instruction::SIToFP:
1409 case Instruction::BitCast:
1410 case Instruction::ZExt:
1411 case Instruction::SExt:
1412 // If the cast is not actually changing bits, and the second operand is a
1413 // null pointer, do the comparison with the pre-casted value.
1414 if (V2->isNullValue() &&
1415 (isa<PointerType>(CE1->getType()) || CE1->getType()->isInteger())) {
1416 if (CE1->getOpcode() == Instruction::ZExt) isSigned = false;
1417 if (CE1->getOpcode() == Instruction::SExt) isSigned = true;
1418 return evaluateICmpRelation(Context, CE1Op0,
1419 Constant::getNullValue(CE1Op0->getType()),
1424 case Instruction::GetElementPtr:
1425 // Ok, since this is a getelementptr, we know that the constant has a
1426 // pointer type. Check the various cases.
1427 if (isa<ConstantPointerNull>(V2)) {
1428 // If we are comparing a GEP to a null pointer, check to see if the base
1429 // of the GEP equals the null pointer.
1430 if (const GlobalValue *GV = dyn_cast<GlobalValue>(CE1Op0)) {
1431 if (GV->hasExternalWeakLinkage())
1432 // Weak linkage GVals could be zero or not. We're comparing that
1433 // to null pointer so its greater-or-equal
1434 return isSigned ? ICmpInst::ICMP_SGE : ICmpInst::ICMP_UGE;
1436 // If its not weak linkage, the GVal must have a non-zero address
1437 // so the result is greater-than
1438 return isSigned ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT;
1439 } else if (isa<ConstantPointerNull>(CE1Op0)) {
1440 // If we are indexing from a null pointer, check to see if we have any
1441 // non-zero indices.
1442 for (unsigned i = 1, e = CE1->getNumOperands(); i != e; ++i)
1443 if (!CE1->getOperand(i)->isNullValue())
1444 // Offsetting from null, must not be equal.
1445 return isSigned ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT;
1446 // Only zero indexes from null, must still be zero.
1447 return ICmpInst::ICMP_EQ;
1449 // Otherwise, we can't really say if the first operand is null or not.
1450 } else if (const GlobalValue *CPR2 = dyn_cast<GlobalValue>(V2)) {
1451 if (isa<ConstantPointerNull>(CE1Op0)) {
1452 if (CPR2->hasExternalWeakLinkage())
1453 // Weak linkage GVals could be zero or not. We're comparing it to
1454 // a null pointer, so its less-or-equal
1455 return isSigned ? ICmpInst::ICMP_SLE : ICmpInst::ICMP_ULE;
1457 // If its not weak linkage, the GVal must have a non-zero address
1458 // so the result is less-than
1459 return isSigned ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT;
1460 } else if (const GlobalValue *CPR1 = dyn_cast<GlobalValue>(CE1Op0)) {
1462 // If this is a getelementptr of the same global, then it must be
1463 // different. Because the types must match, the getelementptr could
1464 // only have at most one index, and because we fold getelementptr's
1465 // with a single zero index, it must be nonzero.
1466 assert(CE1->getNumOperands() == 2 &&
1467 !CE1->getOperand(1)->isNullValue() &&
1468 "Suprising getelementptr!");
1469 return isSigned ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT;
1471 // If they are different globals, we don't know what the value is,
1472 // but they can't be equal.
1473 return ICmpInst::ICMP_NE;
1477 ConstantExpr *CE2 = cast<ConstantExpr>(V2);
1478 Constant *CE2Op0 = CE2->getOperand(0);
1480 // There are MANY other foldings that we could perform here. They will
1481 // probably be added on demand, as they seem needed.
1482 switch (CE2->getOpcode()) {
1484 case Instruction::GetElementPtr:
1485 // By far the most common case to handle is when the base pointers are
1486 // obviously to the same or different globals.
1487 if (isa<GlobalValue>(CE1Op0) && isa<GlobalValue>(CE2Op0)) {
1488 if (CE1Op0 != CE2Op0) // Don't know relative ordering, but not equal
1489 return ICmpInst::ICMP_NE;
1490 // Ok, we know that both getelementptr instructions are based on the
1491 // same global. From this, we can precisely determine the relative
1492 // ordering of the resultant pointers.
1495 // The logic below assumes that the result of the comparison
1496 // can be determined by finding the first index that differs.
1497 // This doesn't work if there is over-indexing in any
1498 // subsequent indices, so check for that case first.
1499 if (!CE1->isGEPWithNoNotionalOverIndexing() ||
1500 !CE2->isGEPWithNoNotionalOverIndexing())
1501 return ICmpInst::BAD_ICMP_PREDICATE; // Might be equal.
1503 // Compare all of the operands the GEP's have in common.
1504 gep_type_iterator GTI = gep_type_begin(CE1);
1505 for (;i != CE1->getNumOperands() && i != CE2->getNumOperands();
1507 switch (IdxCompare(Context, CE1->getOperand(i),
1508 CE2->getOperand(i), GTI.getIndexedType())) {
1509 case -1: return isSigned ? ICmpInst::ICMP_SLT:ICmpInst::ICMP_ULT;
1510 case 1: return isSigned ? ICmpInst::ICMP_SGT:ICmpInst::ICMP_UGT;
1511 case -2: return ICmpInst::BAD_ICMP_PREDICATE;
1514 // Ok, we ran out of things they have in common. If any leftovers
1515 // are non-zero then we have a difference, otherwise we are equal.
1516 for (; i < CE1->getNumOperands(); ++i)
1517 if (!CE1->getOperand(i)->isNullValue()) {
1518 if (isa<ConstantInt>(CE1->getOperand(i)))
1519 return isSigned ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT;
1521 return ICmpInst::BAD_ICMP_PREDICATE; // Might be equal.
1524 for (; i < CE2->getNumOperands(); ++i)
1525 if (!CE2->getOperand(i)->isNullValue()) {
1526 if (isa<ConstantInt>(CE2->getOperand(i)))
1527 return isSigned ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT;
1529 return ICmpInst::BAD_ICMP_PREDICATE; // Might be equal.
1531 return ICmpInst::ICMP_EQ;
1540 return ICmpInst::BAD_ICMP_PREDICATE;
1543 Constant *llvm::ConstantFoldCompareInstruction(LLVMContext &Context,
1544 unsigned short pred,
1545 Constant *C1, Constant *C2) {
1546 const Type *ResultTy;
1547 if (const VectorType *VT = dyn_cast<VectorType>(C1->getType()))
1548 ResultTy = VectorType::get(Type::getInt1Ty(Context), VT->getNumElements());
1550 ResultTy = Type::getInt1Ty(Context);
1552 // Fold FCMP_FALSE/FCMP_TRUE unconditionally.
1553 if (pred == FCmpInst::FCMP_FALSE)
1554 return Constant::getNullValue(ResultTy);
1556 if (pred == FCmpInst::FCMP_TRUE)
1557 return Constant::getAllOnesValue(ResultTy);
1559 // Handle some degenerate cases first
1560 if (isa<UndefValue>(C1) || isa<UndefValue>(C2))
1561 return UndefValue::get(ResultTy);
1563 // No compile-time operations on this type yet.
1564 if (C1->getType()->isPPC_FP128Ty())
1567 // icmp eq/ne(null,GV) -> false/true
1568 if (C1->isNullValue()) {
1569 if (const GlobalValue *GV = dyn_cast<GlobalValue>(C2))
1570 // Don't try to evaluate aliases. External weak GV can be null.
1571 if (!isa<GlobalAlias>(GV) && !GV->hasExternalWeakLinkage()) {
1572 if (pred == ICmpInst::ICMP_EQ)
1573 return ConstantInt::getFalse(Context);
1574 else if (pred == ICmpInst::ICMP_NE)
1575 return ConstantInt::getTrue(Context);
1577 // icmp eq/ne(GV,null) -> false/true
1578 } else if (C2->isNullValue()) {
1579 if (const GlobalValue *GV = dyn_cast<GlobalValue>(C1))
1580 // Don't try to evaluate aliases. External weak GV can be null.
1581 if (!isa<GlobalAlias>(GV) && !GV->hasExternalWeakLinkage()) {
1582 if (pred == ICmpInst::ICMP_EQ)
1583 return ConstantInt::getFalse(Context);
1584 else if (pred == ICmpInst::ICMP_NE)
1585 return ConstantInt::getTrue(Context);
1589 // If the comparison is a comparison between two i1's, simplify it.
1590 if (C1->getType() == Type::getInt1Ty(Context)) {
1592 case ICmpInst::ICMP_EQ:
1593 if (isa<ConstantInt>(C2))
1594 return ConstantExpr::getXor(C1, ConstantExpr::getNot(C2));
1595 return ConstantExpr::getXor(ConstantExpr::getNot(C1), C2);
1596 case ICmpInst::ICMP_NE:
1597 return ConstantExpr::getXor(C1, C2);
1603 if (isa<ConstantInt>(C1) && isa<ConstantInt>(C2)) {
1604 APInt V1 = cast<ConstantInt>(C1)->getValue();
1605 APInt V2 = cast<ConstantInt>(C2)->getValue();
1607 default: llvm_unreachable("Invalid ICmp Predicate"); return 0;
1608 case ICmpInst::ICMP_EQ:
1609 return ConstantInt::get(Type::getInt1Ty(Context), V1 == V2);
1610 case ICmpInst::ICMP_NE:
1611 return ConstantInt::get(Type::getInt1Ty(Context), V1 != V2);
1612 case ICmpInst::ICMP_SLT:
1613 return ConstantInt::get(Type::getInt1Ty(Context), V1.slt(V2));
1614 case ICmpInst::ICMP_SGT:
1615 return ConstantInt::get(Type::getInt1Ty(Context), V1.sgt(V2));
1616 case ICmpInst::ICMP_SLE:
1617 return ConstantInt::get(Type::getInt1Ty(Context), V1.sle(V2));
1618 case ICmpInst::ICMP_SGE:
1619 return ConstantInt::get(Type::getInt1Ty(Context), V1.sge(V2));
1620 case ICmpInst::ICMP_ULT:
1621 return ConstantInt::get(Type::getInt1Ty(Context), V1.ult(V2));
1622 case ICmpInst::ICMP_UGT:
1623 return ConstantInt::get(Type::getInt1Ty(Context), V1.ugt(V2));
1624 case ICmpInst::ICMP_ULE:
1625 return ConstantInt::get(Type::getInt1Ty(Context), V1.ule(V2));
1626 case ICmpInst::ICMP_UGE:
1627 return ConstantInt::get(Type::getInt1Ty(Context), V1.uge(V2));
1629 } else if (isa<ConstantFP>(C1) && isa<ConstantFP>(C2)) {
1630 APFloat C1V = cast<ConstantFP>(C1)->getValueAPF();
1631 APFloat C2V = cast<ConstantFP>(C2)->getValueAPF();
1632 APFloat::cmpResult R = C1V.compare(C2V);
1634 default: llvm_unreachable("Invalid FCmp Predicate"); return 0;
1635 case FCmpInst::FCMP_FALSE: return ConstantInt::getFalse(Context);
1636 case FCmpInst::FCMP_TRUE: return ConstantInt::getTrue(Context);
1637 case FCmpInst::FCMP_UNO:
1638 return ConstantInt::get(Type::getInt1Ty(Context), R==APFloat::cmpUnordered);
1639 case FCmpInst::FCMP_ORD:
1640 return ConstantInt::get(Type::getInt1Ty(Context), R!=APFloat::cmpUnordered);
1641 case FCmpInst::FCMP_UEQ:
1642 return ConstantInt::get(Type::getInt1Ty(Context), R==APFloat::cmpUnordered ||
1643 R==APFloat::cmpEqual);
1644 case FCmpInst::FCMP_OEQ:
1645 return ConstantInt::get(Type::getInt1Ty(Context), R==APFloat::cmpEqual);
1646 case FCmpInst::FCMP_UNE:
1647 return ConstantInt::get(Type::getInt1Ty(Context), R!=APFloat::cmpEqual);
1648 case FCmpInst::FCMP_ONE:
1649 return ConstantInt::get(Type::getInt1Ty(Context), R==APFloat::cmpLessThan ||
1650 R==APFloat::cmpGreaterThan);
1651 case FCmpInst::FCMP_ULT:
1652 return ConstantInt::get(Type::getInt1Ty(Context), R==APFloat::cmpUnordered ||
1653 R==APFloat::cmpLessThan);
1654 case FCmpInst::FCMP_OLT:
1655 return ConstantInt::get(Type::getInt1Ty(Context), R==APFloat::cmpLessThan);
1656 case FCmpInst::FCMP_UGT:
1657 return ConstantInt::get(Type::getInt1Ty(Context), R==APFloat::cmpUnordered ||
1658 R==APFloat::cmpGreaterThan);
1659 case FCmpInst::FCMP_OGT:
1660 return ConstantInt::get(Type::getInt1Ty(Context), R==APFloat::cmpGreaterThan);
1661 case FCmpInst::FCMP_ULE:
1662 return ConstantInt::get(Type::getInt1Ty(Context), R!=APFloat::cmpGreaterThan);
1663 case FCmpInst::FCMP_OLE:
1664 return ConstantInt::get(Type::getInt1Ty(Context), R==APFloat::cmpLessThan ||
1665 R==APFloat::cmpEqual);
1666 case FCmpInst::FCMP_UGE:
1667 return ConstantInt::get(Type::getInt1Ty(Context), R!=APFloat::cmpLessThan);
1668 case FCmpInst::FCMP_OGE:
1669 return ConstantInt::get(Type::getInt1Ty(Context), R==APFloat::cmpGreaterThan ||
1670 R==APFloat::cmpEqual);
1672 } else if (isa<VectorType>(C1->getType())) {
1673 SmallVector<Constant*, 16> C1Elts, C2Elts;
1674 C1->getVectorElements(Context, C1Elts);
1675 C2->getVectorElements(Context, C2Elts);
1677 // If we can constant fold the comparison of each element, constant fold
1678 // the whole vector comparison.
1679 SmallVector<Constant*, 4> ResElts;
1680 for (unsigned i = 0, e = C1Elts.size(); i != e; ++i) {
1681 // Compare the elements, producing an i1 result or constant expr.
1683 ConstantExpr::getCompare(pred, C1Elts[i], C2Elts[i]));
1685 return ConstantVector::get(&ResElts[0], ResElts.size());
1688 if (C1->getType()->isFloatingPoint()) {
1689 int Result = -1; // -1 = unknown, 0 = known false, 1 = known true.
1690 switch (evaluateFCmpRelation(Context, C1, C2)) {
1691 default: llvm_unreachable("Unknown relation!");
1692 case FCmpInst::FCMP_UNO:
1693 case FCmpInst::FCMP_ORD:
1694 case FCmpInst::FCMP_UEQ:
1695 case FCmpInst::FCMP_UNE:
1696 case FCmpInst::FCMP_ULT:
1697 case FCmpInst::FCMP_UGT:
1698 case FCmpInst::FCMP_ULE:
1699 case FCmpInst::FCMP_UGE:
1700 case FCmpInst::FCMP_TRUE:
1701 case FCmpInst::FCMP_FALSE:
1702 case FCmpInst::BAD_FCMP_PREDICATE:
1703 break; // Couldn't determine anything about these constants.
1704 case FCmpInst::FCMP_OEQ: // We know that C1 == C2
1705 Result = (pred == FCmpInst::FCMP_UEQ || pred == FCmpInst::FCMP_OEQ ||
1706 pred == FCmpInst::FCMP_ULE || pred == FCmpInst::FCMP_OLE ||
1707 pred == FCmpInst::FCMP_UGE || pred == FCmpInst::FCMP_OGE);
1709 case FCmpInst::FCMP_OLT: // We know that C1 < C2
1710 Result = (pred == FCmpInst::FCMP_UNE || pred == FCmpInst::FCMP_ONE ||
1711 pred == FCmpInst::FCMP_ULT || pred == FCmpInst::FCMP_OLT ||
1712 pred == FCmpInst::FCMP_ULE || pred == FCmpInst::FCMP_OLE);
1714 case FCmpInst::FCMP_OGT: // We know that C1 > C2
1715 Result = (pred == FCmpInst::FCMP_UNE || pred == FCmpInst::FCMP_ONE ||
1716 pred == FCmpInst::FCMP_UGT || pred == FCmpInst::FCMP_OGT ||
1717 pred == FCmpInst::FCMP_UGE || pred == FCmpInst::FCMP_OGE);
1719 case FCmpInst::FCMP_OLE: // We know that C1 <= C2
1720 // We can only partially decide this relation.
1721 if (pred == FCmpInst::FCMP_UGT || pred == FCmpInst::FCMP_OGT)
1723 else if (pred == FCmpInst::FCMP_ULT || pred == FCmpInst::FCMP_OLT)
1726 case FCmpInst::FCMP_OGE: // We known that C1 >= C2
1727 // We can only partially decide this relation.
1728 if (pred == FCmpInst::FCMP_ULT || pred == FCmpInst::FCMP_OLT)
1730 else if (pred == FCmpInst::FCMP_UGT || pred == FCmpInst::FCMP_OGT)
1733 case ICmpInst::ICMP_NE: // We know that C1 != C2
1734 // We can only partially decide this relation.
1735 if (pred == FCmpInst::FCMP_OEQ || pred == FCmpInst::FCMP_UEQ)
1737 else if (pred == FCmpInst::FCMP_ONE || pred == FCmpInst::FCMP_UNE)
1742 // If we evaluated the result, return it now.
1744 return ConstantInt::get(Type::getInt1Ty(Context), Result);
1747 // Evaluate the relation between the two constants, per the predicate.
1748 int Result = -1; // -1 = unknown, 0 = known false, 1 = known true.
1749 switch (evaluateICmpRelation(Context, C1, C2, CmpInst::isSigned(pred))) {
1750 default: llvm_unreachable("Unknown relational!");
1751 case ICmpInst::BAD_ICMP_PREDICATE:
1752 break; // Couldn't determine anything about these constants.
1753 case ICmpInst::ICMP_EQ: // We know the constants are equal!
1754 // If we know the constants are equal, we can decide the result of this
1755 // computation precisely.
1756 Result = ICmpInst::isTrueWhenEqual((ICmpInst::Predicate)pred);
1758 case ICmpInst::ICMP_ULT:
1760 case ICmpInst::ICMP_ULT: case ICmpInst::ICMP_NE: case ICmpInst::ICMP_ULE:
1762 case ICmpInst::ICMP_UGT: case ICmpInst::ICMP_EQ: case ICmpInst::ICMP_UGE:
1766 case ICmpInst::ICMP_SLT:
1768 case ICmpInst::ICMP_SLT: case ICmpInst::ICMP_NE: case ICmpInst::ICMP_SLE:
1770 case ICmpInst::ICMP_SGT: case ICmpInst::ICMP_EQ: case ICmpInst::ICMP_SGE:
1774 case ICmpInst::ICMP_UGT:
1776 case ICmpInst::ICMP_UGT: case ICmpInst::ICMP_NE: case ICmpInst::ICMP_UGE:
1778 case ICmpInst::ICMP_ULT: case ICmpInst::ICMP_EQ: case ICmpInst::ICMP_ULE:
1782 case ICmpInst::ICMP_SGT:
1784 case ICmpInst::ICMP_SGT: case ICmpInst::ICMP_NE: case ICmpInst::ICMP_SGE:
1786 case ICmpInst::ICMP_SLT: case ICmpInst::ICMP_EQ: case ICmpInst::ICMP_SLE:
1790 case ICmpInst::ICMP_ULE:
1791 if (pred == ICmpInst::ICMP_UGT) Result = 0;
1792 if (pred == ICmpInst::ICMP_ULT || pred == ICmpInst::ICMP_ULE) Result = 1;
1794 case ICmpInst::ICMP_SLE:
1795 if (pred == ICmpInst::ICMP_SGT) Result = 0;
1796 if (pred == ICmpInst::ICMP_SLT || pred == ICmpInst::ICMP_SLE) Result = 1;
1798 case ICmpInst::ICMP_UGE:
1799 if (pred == ICmpInst::ICMP_ULT) Result = 0;
1800 if (pred == ICmpInst::ICMP_UGT || pred == ICmpInst::ICMP_UGE) Result = 1;
1802 case ICmpInst::ICMP_SGE:
1803 if (pred == ICmpInst::ICMP_SLT) Result = 0;
1804 if (pred == ICmpInst::ICMP_SGT || pred == ICmpInst::ICMP_SGE) Result = 1;
1806 case ICmpInst::ICMP_NE:
1807 if (pred == ICmpInst::ICMP_EQ) Result = 0;
1808 if (pred == ICmpInst::ICMP_NE) Result = 1;
1812 // If we evaluated the result, return it now.
1814 return ConstantInt::get(Type::getInt1Ty(Context), Result);
1816 // If the right hand side is a bitcast, try using its inverse to simplify
1817 // it by moving it to the left hand side.
1818 if (ConstantExpr *CE2 = dyn_cast<ConstantExpr>(C2)) {
1819 if (CE2->getOpcode() == Instruction::BitCast) {
1820 Constant *CE2Op0 = CE2->getOperand(0);
1821 Constant *Inverse = ConstantExpr::getBitCast(C1, CE2Op0->getType());
1822 return ConstantExpr::getICmp(pred, Inverse, CE2Op0);
1826 // If the left hand side is an extension, try eliminating it.
1827 if (ConstantExpr *CE1 = dyn_cast<ConstantExpr>(C1)) {
1828 if (CE1->getOpcode() == Instruction::SExt ||
1829 CE1->getOpcode() == Instruction::ZExt) {
1830 Constant *CE1Op0 = CE1->getOperand(0);
1831 Constant *CE1Inverse = ConstantExpr::getTrunc(CE1, CE1Op0->getType());
1832 if (CE1Inverse == CE1Op0) {
1833 // Check whether we can safely truncate the right hand side.
1834 Constant *C2Inverse = ConstantExpr::getTrunc(C2, CE1Op0->getType());
1835 if (ConstantExpr::getZExt(C2Inverse, C2->getType()) == C2) {
1836 return ConstantExpr::getICmp(pred, CE1Inverse, C2Inverse);
1842 if (!isa<ConstantExpr>(C1) && isa<ConstantExpr>(C2)) {
1843 // If C2 is a constant expr and C1 isn't, flip them around and fold the
1844 // other way if possible.
1846 case ICmpInst::ICMP_EQ:
1847 case ICmpInst::ICMP_NE:
1848 // No change of predicate required.
1849 return ConstantFoldCompareInstruction(Context, pred, C2, C1);
1851 case ICmpInst::ICMP_ULT:
1852 case ICmpInst::ICMP_SLT:
1853 case ICmpInst::ICMP_UGT:
1854 case ICmpInst::ICMP_SGT:
1855 case ICmpInst::ICMP_ULE:
1856 case ICmpInst::ICMP_SLE:
1857 case ICmpInst::ICMP_UGE:
1858 case ICmpInst::ICMP_SGE:
1859 // Change the predicate as necessary to swap the operands.
1860 pred = ICmpInst::getSwappedPredicate((ICmpInst::Predicate)pred);
1861 return ConstantFoldCompareInstruction(Context, pred, C2, C1);
1863 default: // These predicates cannot be flopped around.
1871 /// isInBoundsIndices - Test whether the given sequence of *normalized* indices
1873 static bool isInBoundsIndices(Constant *const *Idxs, size_t NumIdx) {
1874 // No indices means nothing that could be out of bounds.
1875 if (NumIdx == 0) return true;
1877 // If the first index is zero, it's in bounds.
1878 if (Idxs[0]->isNullValue()) return true;
1880 // If the first index is one and all the rest are zero, it's in bounds,
1881 // by the one-past-the-end rule.
1882 if (!cast<ConstantInt>(Idxs[0])->isOne())
1884 for (unsigned i = 1, e = NumIdx; i != e; ++i)
1885 if (!Idxs[i]->isNullValue())
1890 Constant *llvm::ConstantFoldGetElementPtr(LLVMContext &Context,
1893 Constant* const *Idxs,
1896 (NumIdx == 1 && Idxs[0]->isNullValue()))
1899 if (isa<UndefValue>(C)) {
1900 const PointerType *Ptr = cast<PointerType>(C->getType());
1901 const Type *Ty = GetElementPtrInst::getIndexedType(Ptr,
1903 (Value **)Idxs+NumIdx);
1904 assert(Ty != 0 && "Invalid indices for GEP!");
1905 return UndefValue::get(PointerType::get(Ty, Ptr->getAddressSpace()));
1908 Constant *Idx0 = Idxs[0];
1909 if (C->isNullValue()) {
1911 for (unsigned i = 0, e = NumIdx; i != e; ++i)
1912 if (!Idxs[i]->isNullValue()) {
1917 const PointerType *Ptr = cast<PointerType>(C->getType());
1918 const Type *Ty = GetElementPtrInst::getIndexedType(Ptr,
1920 (Value**)Idxs+NumIdx);
1921 assert(Ty != 0 && "Invalid indices for GEP!");
1922 return ConstantPointerNull::get(
1923 PointerType::get(Ty,Ptr->getAddressSpace()));
1927 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C)) {
1928 // Combine Indices - If the source pointer to this getelementptr instruction
1929 // is a getelementptr instruction, combine the indices of the two
1930 // getelementptr instructions into a single instruction.
1932 if (CE->getOpcode() == Instruction::GetElementPtr) {
1933 const Type *LastTy = 0;
1934 for (gep_type_iterator I = gep_type_begin(CE), E = gep_type_end(CE);
1938 if ((LastTy && isa<ArrayType>(LastTy)) || Idx0->isNullValue()) {
1939 SmallVector<Value*, 16> NewIndices;
1940 NewIndices.reserve(NumIdx + CE->getNumOperands());
1941 for (unsigned i = 1, e = CE->getNumOperands()-1; i != e; ++i)
1942 NewIndices.push_back(CE->getOperand(i));
1944 // Add the last index of the source with the first index of the new GEP.
1945 // Make sure to handle the case when they are actually different types.
1946 Constant *Combined = CE->getOperand(CE->getNumOperands()-1);
1947 // Otherwise it must be an array.
1948 if (!Idx0->isNullValue()) {
1949 const Type *IdxTy = Combined->getType();
1950 if (IdxTy != Idx0->getType()) {
1952 ConstantExpr::getSExtOrBitCast(Idx0, Type::getInt64Ty(Context));
1953 Constant *C2 = ConstantExpr::getSExtOrBitCast(Combined,
1954 Type::getInt64Ty(Context));
1955 Combined = ConstantExpr::get(Instruction::Add, C1, C2);
1958 ConstantExpr::get(Instruction::Add, Idx0, Combined);
1962 NewIndices.push_back(Combined);
1963 NewIndices.insert(NewIndices.end(), Idxs+1, Idxs+NumIdx);
1964 return (inBounds && cast<GEPOperator>(CE)->isInBounds()) ?
1965 ConstantExpr::getInBoundsGetElementPtr(CE->getOperand(0),
1967 NewIndices.size()) :
1968 ConstantExpr::getGetElementPtr(CE->getOperand(0),
1974 // Implement folding of:
1975 // int* getelementptr ([2 x int]* cast ([3 x int]* %X to [2 x int]*),
1977 // To: int* getelementptr ([3 x int]* %X, long 0, long 0)
1979 if (CE->isCast() && NumIdx > 1 && Idx0->isNullValue()) {
1980 if (const PointerType *SPT =
1981 dyn_cast<PointerType>(CE->getOperand(0)->getType()))
1982 if (const ArrayType *SAT = dyn_cast<ArrayType>(SPT->getElementType()))
1983 if (const ArrayType *CAT =
1984 dyn_cast<ArrayType>(cast<PointerType>(C->getType())->getElementType()))
1985 if (CAT->getElementType() == SAT->getElementType())
1987 ConstantExpr::getInBoundsGetElementPtr(
1988 (Constant*)CE->getOperand(0), Idxs, NumIdx) :
1989 ConstantExpr::getGetElementPtr(
1990 (Constant*)CE->getOperand(0), Idxs, NumIdx);
1993 // Fold: getelementptr (i8* inttoptr (i64 1 to i8*), i32 -1)
1994 // Into: inttoptr (i64 0 to i8*)
1995 // This happens with pointers to member functions in C++.
1996 if (CE->getOpcode() == Instruction::IntToPtr && NumIdx == 1 &&
1997 isa<ConstantInt>(CE->getOperand(0)) && isa<ConstantInt>(Idxs[0]) &&
1998 cast<PointerType>(CE->getType())->getElementType() ==
1999 Type::getInt8Ty(Context)) {
2000 Constant *Base = CE->getOperand(0);
2001 Constant *Offset = Idxs[0];
2003 // Convert the smaller integer to the larger type.
2004 if (Offset->getType()->getPrimitiveSizeInBits() <
2005 Base->getType()->getPrimitiveSizeInBits())
2006 Offset = ConstantExpr::getSExt(Offset, Base->getType());
2007 else if (Base->getType()->getPrimitiveSizeInBits() <
2008 Offset->getType()->getPrimitiveSizeInBits())
2009 Base = ConstantExpr::getZExt(Base, Offset->getType());
2011 Base = ConstantExpr::getAdd(Base, Offset);
2012 return ConstantExpr::getIntToPtr(Base, CE->getType());
2016 // Check to see if any array indices are not within the corresponding
2017 // notional array bounds. If so, try to determine if they can be factored
2018 // out into preceding dimensions.
2019 bool Unknown = false;
2020 SmallVector<Constant *, 8> NewIdxs;
2021 const Type *Ty = C->getType();
2022 const Type *Prev = 0;
2023 for (unsigned i = 0; i != NumIdx;
2024 Prev = Ty, Ty = cast<CompositeType>(Ty)->getTypeAtIndex(Idxs[i]), ++i) {
2025 if (ConstantInt *CI = dyn_cast<ConstantInt>(Idxs[i])) {
2026 if (const ArrayType *ATy = dyn_cast<ArrayType>(Ty))
2027 if (ATy->getNumElements() <= INT64_MAX &&
2028 ATy->getNumElements() != 0 &&
2029 CI->getSExtValue() >= (int64_t)ATy->getNumElements()) {
2030 if (isa<SequentialType>(Prev)) {
2031 // It's out of range, but we can factor it into the prior
2033 NewIdxs.resize(NumIdx);
2034 ConstantInt *Factor = ConstantInt::get(CI->getType(),
2035 ATy->getNumElements());
2036 NewIdxs[i] = ConstantExpr::getSRem(CI, Factor);
2038 Constant *PrevIdx = Idxs[i-1];
2039 Constant *Div = ConstantExpr::getSDiv(CI, Factor);
2041 // Before adding, extend both operands to i64 to avoid
2042 // overflow trouble.
2043 if (PrevIdx->getType() != Type::getInt64Ty(Context))
2044 PrevIdx = ConstantExpr::getSExt(PrevIdx,
2045 Type::getInt64Ty(Context));
2046 if (Div->getType() != Type::getInt64Ty(Context))
2047 Div = ConstantExpr::getSExt(Div,
2048 Type::getInt64Ty(Context));
2050 NewIdxs[i-1] = ConstantExpr::getAdd(PrevIdx, Div);
2052 // It's out of range, but the prior dimension is a struct
2053 // so we can't do anything about it.
2058 // We don't know if it's in range or not.
2063 // If we did any factoring, start over with the adjusted indices.
2064 if (!NewIdxs.empty()) {
2065 for (unsigned i = 0; i != NumIdx; ++i)
2066 if (!NewIdxs[i]) NewIdxs[i] = Idxs[i];
2068 ConstantExpr::getInBoundsGetElementPtr(C, NewIdxs.data(),
2070 ConstantExpr::getGetElementPtr(C, NewIdxs.data(), NewIdxs.size());
2073 // If all indices are known integers and normalized, we can do a simple
2074 // check for the "inbounds" property.
2075 if (!Unknown && !inBounds &&
2076 isa<GlobalVariable>(C) && isInBoundsIndices(Idxs, NumIdx))
2077 return ConstantExpr::getInBoundsGetElementPtr(C, Idxs, NumIdx);