add a version of the APFloat constructor that initializes to 0.0
[llvm/avr.git] / lib / VMCore / ConstantFold.cpp
blob15b6df90a44da6121ecba074fe2e3c9511e29371
1 //===- ConstantFold.cpp - LLVM constant folder ----------------------------===//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file implements 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"
34 #include <limits>
35 using namespace llvm;
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())
51 return 0;
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)))
57 return 0;
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),
65 DstEltTy));
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
73 static unsigned
74 foldConstantCastPair(
75 unsigned opc, ///< opcode of the second cast constant expression
76 const ConstantExpr*Op, ///< the first cast constant expression
77 const Type *DstTy ///< desintation type of the first cast
78 ) {
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();
97 if (SrcTy == DestTy)
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);
119 } else {
120 break;
124 if (ElTy == DPTy->getElementType())
125 // This GEP is inbounds because all indices are zero.
126 return ConstantExpr::getInBoundsGetElementPtr(V, &IdxList[0],
127 IdxList.size());
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!");
136 SrcTy = NULL;
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
154 // bitcast right.
155 if (isa<ConstantPointerNull>(V)) // ptr->ptr cast.
156 return ConstantPointerNull::get(cast<PointerType>(DestTy));
158 // Handle integral constant input.
159 if (const 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.
163 return 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?)
170 return 0;
173 // Handle ConstantFP input.
174 if (const ConstantFP *FP = dyn_cast<ConstantFP>(V))
175 // FP -> Integral.
176 return ConstantInt::get(Context, FP->getValueAPF().bitcastToAPInt());
178 return 0;
182 Constant *llvm::ConstantFoldCastInstruction(LLVMContext &Context,
183 unsigned opc, const Constant *V,
184 const Type *DestTy) {
185 if (isa<UndefValue>(V)) {
186 // zext(undef) = 0, because the top bits will be zero.
187 // sext(undef) = 0, because the top bits will all be the same.
188 // [us]itofp(undef) = 0, because the result value is bounded.
189 if (opc == Instruction::ZExt || opc == Instruction::SExt ||
190 opc == Instruction::UIToFP || opc == Instruction::SIToFP)
191 return Constant::getNullValue(DestTy);
192 return UndefValue::get(DestTy);
194 // No compile-time operations on this type yet.
195 if (V->getType() == Type::getPPC_FP128Ty(Context) || DestTy == Type::getPPC_FP128Ty(Context))
196 return 0;
198 // If the cast operand is a constant expression, there's a few things we can
199 // do to try to simplify it.
200 if (const ConstantExpr *CE = dyn_cast<ConstantExpr>(V)) {
201 if (CE->isCast()) {
202 // Try hard to fold cast of cast because they are often eliminable.
203 if (unsigned newOpc = foldConstantCastPair(opc, CE, DestTy))
204 return ConstantExpr::getCast(newOpc, CE->getOperand(0), DestTy);
205 } else if (CE->getOpcode() == Instruction::GetElementPtr) {
206 // If all of the indexes in the GEP are null values, there is no pointer
207 // adjustment going on. We might as well cast the source pointer.
208 bool isAllNull = true;
209 for (unsigned i = 1, e = CE->getNumOperands(); i != e; ++i)
210 if (!CE->getOperand(i)->isNullValue()) {
211 isAllNull = false;
212 break;
214 if (isAllNull)
215 // This is casting one pointer type to another, always BitCast
216 return ConstantExpr::getPointerCast(CE->getOperand(0), DestTy);
220 // If the cast operand is a constant vector, perform the cast by
221 // operating on each element. In the cast of bitcasts, the element
222 // count may be mismatched; don't attempt to handle that here.
223 if (const ConstantVector *CV = dyn_cast<ConstantVector>(V))
224 if (isa<VectorType>(DestTy) &&
225 cast<VectorType>(DestTy)->getNumElements() ==
226 CV->getType()->getNumElements()) {
227 std::vector<Constant*> res;
228 const VectorType *DestVecTy = cast<VectorType>(DestTy);
229 const Type *DstEltTy = DestVecTy->getElementType();
230 for (unsigned i = 0, e = CV->getType()->getNumElements(); i != e; ++i)
231 res.push_back(ConstantExpr::getCast(opc,
232 CV->getOperand(i), DstEltTy));
233 return ConstantVector::get(DestVecTy, res);
236 // We actually have to do a cast now. Perform the cast according to the
237 // opcode specified.
238 switch (opc) {
239 case Instruction::FPTrunc:
240 case Instruction::FPExt:
241 if (const ConstantFP *FPC = dyn_cast<ConstantFP>(V)) {
242 bool ignored;
243 APFloat Val = FPC->getValueAPF();
244 Val.convert(DestTy == Type::getFloatTy(Context) ? APFloat::IEEEsingle :
245 DestTy == Type::getDoubleTy(Context) ? APFloat::IEEEdouble :
246 DestTy == Type::getX86_FP80Ty(Context) ? APFloat::x87DoubleExtended :
247 DestTy == Type::getFP128Ty(Context) ? APFloat::IEEEquad :
248 APFloat::Bogus,
249 APFloat::rmNearestTiesToEven, &ignored);
250 return ConstantFP::get(Context, Val);
252 return 0; // Can't fold.
253 case Instruction::FPToUI:
254 case Instruction::FPToSI:
255 if (const ConstantFP *FPC = dyn_cast<ConstantFP>(V)) {
256 const APFloat &V = FPC->getValueAPF();
257 bool ignored;
258 uint64_t x[2];
259 uint32_t DestBitWidth = cast<IntegerType>(DestTy)->getBitWidth();
260 (void) V.convertToInteger(x, DestBitWidth, opc==Instruction::FPToSI,
261 APFloat::rmTowardZero, &ignored);
262 APInt Val(DestBitWidth, 2, x);
263 return ConstantInt::get(Context, Val);
265 return 0; // Can't fold.
266 case Instruction::IntToPtr: //always treated as unsigned
267 if (V->isNullValue()) // Is it an integral null value?
268 return ConstantPointerNull::get(cast<PointerType>(DestTy));
269 return 0; // Other pointer types cannot be casted
270 case Instruction::PtrToInt: // always treated as unsigned
271 if (V->isNullValue()) // is it a null pointer value?
272 return ConstantInt::get(DestTy, 0);
273 return 0; // Other pointer types cannot be casted
274 case Instruction::UIToFP:
275 case Instruction::SIToFP:
276 if (const ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
277 APInt api = CI->getValue();
278 const uint64_t zero[] = {0, 0};
279 APFloat apf = APFloat(APInt(DestTy->getPrimitiveSizeInBits(),
280 2, zero));
281 (void)apf.convertFromAPInt(api,
282 opc==Instruction::SIToFP,
283 APFloat::rmNearestTiesToEven);
284 return ConstantFP::get(Context, apf);
286 return 0;
287 case Instruction::ZExt:
288 if (const ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
289 uint32_t BitWidth = cast<IntegerType>(DestTy)->getBitWidth();
290 APInt Result(CI->getValue());
291 Result.zext(BitWidth);
292 return ConstantInt::get(Context, Result);
294 return 0;
295 case Instruction::SExt:
296 if (const ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
297 uint32_t BitWidth = cast<IntegerType>(DestTy)->getBitWidth();
298 APInt Result(CI->getValue());
299 Result.sext(BitWidth);
300 return ConstantInt::get(Context, Result);
302 return 0;
303 case Instruction::Trunc:
304 if (const ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
305 uint32_t BitWidth = cast<IntegerType>(DestTy)->getBitWidth();
306 APInt Result(CI->getValue());
307 Result.trunc(BitWidth);
308 return ConstantInt::get(Context, Result);
310 return 0;
311 case Instruction::BitCast:
312 return FoldBitCast(Context, const_cast<Constant*>(V), DestTy);
313 default:
314 assert(!"Invalid CE CastInst opcode");
315 break;
318 llvm_unreachable("Failed to cast constant expression");
319 return 0;
322 Constant *llvm::ConstantFoldSelectInstruction(LLVMContext&,
323 const Constant *Cond,
324 const Constant *V1,
325 const Constant *V2) {
326 if (const ConstantInt *CB = dyn_cast<ConstantInt>(Cond))
327 return const_cast<Constant*>(CB->getZExtValue() ? V1 : V2);
329 if (isa<UndefValue>(V1)) return const_cast<Constant*>(V2);
330 if (isa<UndefValue>(V2)) return const_cast<Constant*>(V1);
331 if (isa<UndefValue>(Cond)) return const_cast<Constant*>(V1);
332 if (V1 == V2) return const_cast<Constant*>(V1);
333 return 0;
336 Constant *llvm::ConstantFoldExtractElementInstruction(LLVMContext &Context,
337 const Constant *Val,
338 const Constant *Idx) {
339 if (isa<UndefValue>(Val)) // ee(undef, x) -> undef
340 return UndefValue::get(cast<VectorType>(Val->getType())->getElementType());
341 if (Val->isNullValue()) // ee(zero, x) -> zero
342 return Constant::getNullValue(
343 cast<VectorType>(Val->getType())->getElementType());
345 if (const ConstantVector *CVal = dyn_cast<ConstantVector>(Val)) {
346 if (const ConstantInt *CIdx = dyn_cast<ConstantInt>(Idx)) {
347 return CVal->getOperand(CIdx->getZExtValue());
348 } else if (isa<UndefValue>(Idx)) {
349 // ee({w,x,y,z}, undef) -> w (an arbitrary value).
350 return CVal->getOperand(0);
353 return 0;
356 Constant *llvm::ConstantFoldInsertElementInstruction(LLVMContext &Context,
357 const Constant *Val,
358 const Constant *Elt,
359 const Constant *Idx) {
360 const ConstantInt *CIdx = dyn_cast<ConstantInt>(Idx);
361 if (!CIdx) return 0;
362 APInt idxVal = CIdx->getValue();
363 if (isa<UndefValue>(Val)) {
364 // Insertion of scalar constant into vector undef
365 // Optimize away insertion of undef
366 if (isa<UndefValue>(Elt))
367 return const_cast<Constant*>(Val);
368 // Otherwise break the aggregate undef into multiple undefs and do
369 // the insertion
370 unsigned numOps =
371 cast<VectorType>(Val->getType())->getNumElements();
372 std::vector<Constant*> Ops;
373 Ops.reserve(numOps);
374 for (unsigned i = 0; i < numOps; ++i) {
375 const Constant *Op =
376 (idxVal == i) ? Elt : UndefValue::get(Elt->getType());
377 Ops.push_back(const_cast<Constant*>(Op));
379 return ConstantVector::get(Ops);
381 if (isa<ConstantAggregateZero>(Val)) {
382 // Insertion of scalar constant into vector aggregate zero
383 // Optimize away insertion of zero
384 if (Elt->isNullValue())
385 return const_cast<Constant*>(Val);
386 // Otherwise break the aggregate zero into multiple zeros and do
387 // the insertion
388 unsigned numOps =
389 cast<VectorType>(Val->getType())->getNumElements();
390 std::vector<Constant*> Ops;
391 Ops.reserve(numOps);
392 for (unsigned i = 0; i < numOps; ++i) {
393 const Constant *Op =
394 (idxVal == i) ? Elt : Constant::getNullValue(Elt->getType());
395 Ops.push_back(const_cast<Constant*>(Op));
397 return ConstantVector::get(Ops);
399 if (const ConstantVector *CVal = dyn_cast<ConstantVector>(Val)) {
400 // Insertion of scalar constant into vector constant
401 std::vector<Constant*> Ops;
402 Ops.reserve(CVal->getNumOperands());
403 for (unsigned i = 0; i < CVal->getNumOperands(); ++i) {
404 const Constant *Op =
405 (idxVal == i) ? Elt : cast<Constant>(CVal->getOperand(i));
406 Ops.push_back(const_cast<Constant*>(Op));
408 return ConstantVector::get(Ops);
411 return 0;
414 /// GetVectorElement - If C is a ConstantVector, ConstantAggregateZero or Undef
415 /// return the specified element value. Otherwise return null.
416 static Constant *GetVectorElement(LLVMContext &Context, const Constant *C,
417 unsigned EltNo) {
418 if (const ConstantVector *CV = dyn_cast<ConstantVector>(C))
419 return CV->getOperand(EltNo);
421 const Type *EltTy = cast<VectorType>(C->getType())->getElementType();
422 if (isa<ConstantAggregateZero>(C))
423 return Constant::getNullValue(EltTy);
424 if (isa<UndefValue>(C))
425 return UndefValue::get(EltTy);
426 return 0;
429 Constant *llvm::ConstantFoldShuffleVectorInstruction(LLVMContext &Context,
430 const Constant *V1,
431 const Constant *V2,
432 const Constant *Mask) {
433 // Undefined shuffle mask -> undefined value.
434 if (isa<UndefValue>(Mask)) return UndefValue::get(V1->getType());
436 unsigned MaskNumElts = cast<VectorType>(Mask->getType())->getNumElements();
437 unsigned SrcNumElts = cast<VectorType>(V1->getType())->getNumElements();
438 const Type *EltTy = cast<VectorType>(V1->getType())->getElementType();
440 // Loop over the shuffle mask, evaluating each element.
441 SmallVector<Constant*, 32> Result;
442 for (unsigned i = 0; i != MaskNumElts; ++i) {
443 Constant *InElt = GetVectorElement(Context, Mask, i);
444 if (InElt == 0) return 0;
446 if (isa<UndefValue>(InElt))
447 InElt = UndefValue::get(EltTy);
448 else if (ConstantInt *CI = dyn_cast<ConstantInt>(InElt)) {
449 unsigned Elt = CI->getZExtValue();
450 if (Elt >= SrcNumElts*2)
451 InElt = UndefValue::get(EltTy);
452 else if (Elt >= SrcNumElts)
453 InElt = GetVectorElement(Context, V2, Elt - SrcNumElts);
454 else
455 InElt = GetVectorElement(Context, V1, Elt);
456 if (InElt == 0) return 0;
457 } else {
458 // Unknown value.
459 return 0;
461 Result.push_back(InElt);
464 return ConstantVector::get(&Result[0], Result.size());
467 Constant *llvm::ConstantFoldExtractValueInstruction(LLVMContext &Context,
468 const Constant *Agg,
469 const unsigned *Idxs,
470 unsigned NumIdx) {
471 // Base case: no indices, so return the entire value.
472 if (NumIdx == 0)
473 return const_cast<Constant *>(Agg);
475 if (isa<UndefValue>(Agg)) // ev(undef, x) -> undef
476 return UndefValue::get(ExtractValueInst::getIndexedType(Agg->getType(),
477 Idxs,
478 Idxs + NumIdx));
480 if (isa<ConstantAggregateZero>(Agg)) // ev(0, x) -> 0
481 return
482 Constant::getNullValue(ExtractValueInst::getIndexedType(Agg->getType(),
483 Idxs,
484 Idxs + NumIdx));
486 // Otherwise recurse.
487 return ConstantFoldExtractValueInstruction(Context, Agg->getOperand(*Idxs),
488 Idxs+1, NumIdx-1);
491 Constant *llvm::ConstantFoldInsertValueInstruction(LLVMContext &Context,
492 const Constant *Agg,
493 const Constant *Val,
494 const unsigned *Idxs,
495 unsigned NumIdx) {
496 // Base case: no indices, so replace the entire value.
497 if (NumIdx == 0)
498 return const_cast<Constant *>(Val);
500 if (isa<UndefValue>(Agg)) {
501 // Insertion of constant into aggregate undef
502 // Optimize away insertion of undef.
503 if (isa<UndefValue>(Val))
504 return const_cast<Constant*>(Agg);
506 // Otherwise break the aggregate undef into multiple undefs and do
507 // the insertion.
508 const CompositeType *AggTy = cast<CompositeType>(Agg->getType());
509 unsigned numOps;
510 if (const ArrayType *AR = dyn_cast<ArrayType>(AggTy))
511 numOps = AR->getNumElements();
512 else
513 numOps = cast<StructType>(AggTy)->getNumElements();
515 std::vector<Constant*> Ops(numOps);
516 for (unsigned i = 0; i < numOps; ++i) {
517 const Type *MemberTy = AggTy->getTypeAtIndex(i);
518 const Constant *Op =
519 (*Idxs == i) ?
520 ConstantFoldInsertValueInstruction(Context, UndefValue::get(MemberTy),
521 Val, Idxs+1, NumIdx-1) :
522 UndefValue::get(MemberTy);
523 Ops[i] = const_cast<Constant*>(Op);
526 if (const StructType* ST = dyn_cast<StructType>(AggTy))
527 return ConstantStruct::get(Context, Ops, ST->isPacked());
528 return ConstantArray::get(cast<ArrayType>(AggTy), Ops);
531 if (isa<ConstantAggregateZero>(Agg)) {
532 // Insertion of constant into aggregate zero
533 // Optimize away insertion of zero.
534 if (Val->isNullValue())
535 return const_cast<Constant*>(Agg);
537 // Otherwise break the aggregate zero into multiple zeros and do
538 // the insertion.
539 const CompositeType *AggTy = cast<CompositeType>(Agg->getType());
540 unsigned numOps;
541 if (const ArrayType *AR = dyn_cast<ArrayType>(AggTy))
542 numOps = AR->getNumElements();
543 else
544 numOps = cast<StructType>(AggTy)->getNumElements();
546 std::vector<Constant*> Ops(numOps);
547 for (unsigned i = 0; i < numOps; ++i) {
548 const Type *MemberTy = AggTy->getTypeAtIndex(i);
549 const Constant *Op =
550 (*Idxs == i) ?
551 ConstantFoldInsertValueInstruction(Context,
552 Constant::getNullValue(MemberTy),
553 Val, Idxs+1, NumIdx-1) :
554 Constant::getNullValue(MemberTy);
555 Ops[i] = const_cast<Constant*>(Op);
558 if (const StructType* ST = dyn_cast<StructType>(AggTy))
559 return ConstantStruct::get(Context, Ops, ST->isPacked());
560 return ConstantArray::get(cast<ArrayType>(AggTy), Ops);
563 if (isa<ConstantStruct>(Agg) || isa<ConstantArray>(Agg)) {
564 // Insertion of constant into aggregate constant.
565 std::vector<Constant*> Ops(Agg->getNumOperands());
566 for (unsigned i = 0; i < Agg->getNumOperands(); ++i) {
567 const Constant *Op =
568 (*Idxs == i) ?
569 ConstantFoldInsertValueInstruction(Context, Agg->getOperand(i),
570 Val, Idxs+1, NumIdx-1) :
571 Agg->getOperand(i);
572 Ops[i] = const_cast<Constant*>(Op);
575 if (const StructType* ST = dyn_cast<StructType>(Agg->getType()))
576 return ConstantStruct::get(Context, Ops, ST->isPacked());
577 return ConstantArray::get(cast<ArrayType>(Agg->getType()), Ops);
580 return 0;
584 Constant *llvm::ConstantFoldBinaryInstruction(LLVMContext &Context,
585 unsigned Opcode,
586 const Constant *C1,
587 const Constant *C2) {
588 // No compile-time operations on this type yet.
589 if (C1->getType() == Type::getPPC_FP128Ty(Context))
590 return 0;
592 // Handle UndefValue up front.
593 if (isa<UndefValue>(C1) || isa<UndefValue>(C2)) {
594 switch (Opcode) {
595 case Instruction::Xor:
596 if (isa<UndefValue>(C1) && isa<UndefValue>(C2))
597 // Handle undef ^ undef -> 0 special case. This is a common
598 // idiom (misuse).
599 return Constant::getNullValue(C1->getType());
600 // Fallthrough
601 case Instruction::Add:
602 case Instruction::Sub:
603 return UndefValue::get(C1->getType());
604 case Instruction::Mul:
605 case Instruction::And:
606 return Constant::getNullValue(C1->getType());
607 case Instruction::UDiv:
608 case Instruction::SDiv:
609 case Instruction::URem:
610 case Instruction::SRem:
611 if (!isa<UndefValue>(C2)) // undef / X -> 0
612 return Constant::getNullValue(C1->getType());
613 return const_cast<Constant*>(C2); // X / undef -> undef
614 case Instruction::Or: // X | undef -> -1
615 if (const VectorType *PTy = dyn_cast<VectorType>(C1->getType()))
616 return Constant::getAllOnesValue(PTy);
617 return Constant::getAllOnesValue(C1->getType());
618 case Instruction::LShr:
619 if (isa<UndefValue>(C2) && isa<UndefValue>(C1))
620 return const_cast<Constant*>(C1); // undef lshr undef -> undef
621 return Constant::getNullValue(C1->getType()); // X lshr undef -> 0
622 // undef lshr X -> 0
623 case Instruction::AShr:
624 if (!isa<UndefValue>(C2))
625 return const_cast<Constant*>(C1); // undef ashr X --> undef
626 else if (isa<UndefValue>(C1))
627 return const_cast<Constant*>(C1); // undef ashr undef -> undef
628 else
629 return const_cast<Constant*>(C1); // X ashr undef --> X
630 case Instruction::Shl:
631 // undef << X -> 0 or X << undef -> 0
632 return Constant::getNullValue(C1->getType());
636 // Handle simplifications when the RHS is a constant int.
637 if (const ConstantInt *CI2 = dyn_cast<ConstantInt>(C2)) {
638 switch (Opcode) {
639 case Instruction::Add:
640 if (CI2->equalsInt(0)) return const_cast<Constant*>(C1); // X + 0 == X
641 break;
642 case Instruction::Sub:
643 if (CI2->equalsInt(0)) return const_cast<Constant*>(C1); // X - 0 == X
644 break;
645 case Instruction::Mul:
646 if (CI2->equalsInt(0)) return const_cast<Constant*>(C2); // X * 0 == 0
647 if (CI2->equalsInt(1))
648 return const_cast<Constant*>(C1); // X * 1 == X
649 break;
650 case Instruction::UDiv:
651 case Instruction::SDiv:
652 if (CI2->equalsInt(1))
653 return const_cast<Constant*>(C1); // X / 1 == X
654 if (CI2->equalsInt(0))
655 return UndefValue::get(CI2->getType()); // X / 0 == undef
656 break;
657 case Instruction::URem:
658 case Instruction::SRem:
659 if (CI2->equalsInt(1))
660 return Constant::getNullValue(CI2->getType()); // X % 1 == 0
661 if (CI2->equalsInt(0))
662 return UndefValue::get(CI2->getType()); // X % 0 == undef
663 break;
664 case Instruction::And:
665 if (CI2->isZero()) return const_cast<Constant*>(C2); // X & 0 == 0
666 if (CI2->isAllOnesValue())
667 return const_cast<Constant*>(C1); // X & -1 == X
669 if (const ConstantExpr *CE1 = dyn_cast<ConstantExpr>(C1)) {
670 // (zext i32 to i64) & 4294967295 -> (zext i32 to i64)
671 if (CE1->getOpcode() == Instruction::ZExt) {
672 unsigned DstWidth = CI2->getType()->getBitWidth();
673 unsigned SrcWidth =
674 CE1->getOperand(0)->getType()->getPrimitiveSizeInBits();
675 APInt PossiblySetBits(APInt::getLowBitsSet(DstWidth, SrcWidth));
676 if ((PossiblySetBits & CI2->getValue()) == PossiblySetBits)
677 return const_cast<Constant*>(C1);
680 // If and'ing the address of a global with a constant, fold it.
681 if (CE1->getOpcode() == Instruction::PtrToInt &&
682 isa<GlobalValue>(CE1->getOperand(0))) {
683 GlobalValue *GV = cast<GlobalValue>(CE1->getOperand(0));
685 // Functions are at least 4-byte aligned.
686 unsigned GVAlign = GV->getAlignment();
687 if (isa<Function>(GV))
688 GVAlign = std::max(GVAlign, 4U);
690 if (GVAlign > 1) {
691 unsigned DstWidth = CI2->getType()->getBitWidth();
692 unsigned SrcWidth = std::min(DstWidth, Log2_32(GVAlign));
693 APInt BitsNotSet(APInt::getLowBitsSet(DstWidth, SrcWidth));
695 // If checking bits we know are clear, return zero.
696 if ((CI2->getValue() & BitsNotSet) == CI2->getValue())
697 return Constant::getNullValue(CI2->getType());
701 break;
702 case Instruction::Or:
703 if (CI2->equalsInt(0)) return const_cast<Constant*>(C1); // X | 0 == X
704 if (CI2->isAllOnesValue())
705 return const_cast<Constant*>(C2); // X | -1 == -1
706 break;
707 case Instruction::Xor:
708 if (CI2->equalsInt(0)) return const_cast<Constant*>(C1); // X ^ 0 == X
709 break;
710 case Instruction::AShr:
711 // ashr (zext C to Ty), C2 -> lshr (zext C, CSA), C2
712 if (const ConstantExpr *CE1 = dyn_cast<ConstantExpr>(C1))
713 if (CE1->getOpcode() == Instruction::ZExt) // Top bits known zero.
714 return ConstantExpr::getLShr(const_cast<Constant*>(C1),
715 const_cast<Constant*>(C2));
716 break;
720 // At this point we know neither constant is an UndefValue.
721 if (const ConstantInt *CI1 = dyn_cast<ConstantInt>(C1)) {
722 if (const ConstantInt *CI2 = dyn_cast<ConstantInt>(C2)) {
723 using namespace APIntOps;
724 const APInt &C1V = CI1->getValue();
725 const APInt &C2V = CI2->getValue();
726 switch (Opcode) {
727 default:
728 break;
729 case Instruction::Add:
730 return ConstantInt::get(Context, C1V + C2V);
731 case Instruction::Sub:
732 return ConstantInt::get(Context, C1V - C2V);
733 case Instruction::Mul:
734 return ConstantInt::get(Context, C1V * C2V);
735 case Instruction::UDiv:
736 assert(!CI2->isNullValue() && "Div by zero handled above");
737 return ConstantInt::get(Context, C1V.udiv(C2V));
738 case Instruction::SDiv:
739 assert(!CI2->isNullValue() && "Div by zero handled above");
740 if (C2V.isAllOnesValue() && C1V.isMinSignedValue())
741 return UndefValue::get(CI1->getType()); // MIN_INT / -1 -> undef
742 return ConstantInt::get(Context, C1V.sdiv(C2V));
743 case Instruction::URem:
744 assert(!CI2->isNullValue() && "Div by zero handled above");
745 return ConstantInt::get(Context, C1V.urem(C2V));
746 case Instruction::SRem:
747 assert(!CI2->isNullValue() && "Div by zero handled above");
748 if (C2V.isAllOnesValue() && C1V.isMinSignedValue())
749 return UndefValue::get(CI1->getType()); // MIN_INT % -1 -> undef
750 return ConstantInt::get(Context, C1V.srem(C2V));
751 case Instruction::And:
752 return ConstantInt::get(Context, C1V & C2V);
753 case Instruction::Or:
754 return ConstantInt::get(Context, C1V | C2V);
755 case Instruction::Xor:
756 return ConstantInt::get(Context, C1V ^ C2V);
757 case Instruction::Shl: {
758 uint32_t shiftAmt = C2V.getZExtValue();
759 if (shiftAmt < C1V.getBitWidth())
760 return ConstantInt::get(Context, C1V.shl(shiftAmt));
761 else
762 return UndefValue::get(C1->getType()); // too big shift is undef
764 case Instruction::LShr: {
765 uint32_t shiftAmt = C2V.getZExtValue();
766 if (shiftAmt < C1V.getBitWidth())
767 return ConstantInt::get(Context, C1V.lshr(shiftAmt));
768 else
769 return UndefValue::get(C1->getType()); // too big shift is undef
771 case Instruction::AShr: {
772 uint32_t shiftAmt = C2V.getZExtValue();
773 if (shiftAmt < C1V.getBitWidth())
774 return ConstantInt::get(Context, C1V.ashr(shiftAmt));
775 else
776 return UndefValue::get(C1->getType()); // too big shift is undef
781 switch (Opcode) {
782 case Instruction::SDiv:
783 case Instruction::UDiv:
784 case Instruction::URem:
785 case Instruction::SRem:
786 case Instruction::LShr:
787 case Instruction::AShr:
788 case Instruction::Shl:
789 if (CI1->equalsInt(0)) return const_cast<Constant*>(C1);
790 break;
791 default:
792 break;
794 } else if (const ConstantFP *CFP1 = dyn_cast<ConstantFP>(C1)) {
795 if (const ConstantFP *CFP2 = dyn_cast<ConstantFP>(C2)) {
796 APFloat C1V = CFP1->getValueAPF();
797 APFloat C2V = CFP2->getValueAPF();
798 APFloat C3V = C1V; // copy for modification
799 switch (Opcode) {
800 default:
801 break;
802 case Instruction::FAdd:
803 (void)C3V.add(C2V, APFloat::rmNearestTiesToEven);
804 return ConstantFP::get(Context, C3V);
805 case Instruction::FSub:
806 (void)C3V.subtract(C2V, APFloat::rmNearestTiesToEven);
807 return ConstantFP::get(Context, C3V);
808 case Instruction::FMul:
809 (void)C3V.multiply(C2V, APFloat::rmNearestTiesToEven);
810 return ConstantFP::get(Context, C3V);
811 case Instruction::FDiv:
812 (void)C3V.divide(C2V, APFloat::rmNearestTiesToEven);
813 return ConstantFP::get(Context, C3V);
814 case Instruction::FRem:
815 (void)C3V.mod(C2V, APFloat::rmNearestTiesToEven);
816 return ConstantFP::get(Context, C3V);
819 } else if (const VectorType *VTy = dyn_cast<VectorType>(C1->getType())) {
820 const ConstantVector *CP1 = dyn_cast<ConstantVector>(C1);
821 const ConstantVector *CP2 = dyn_cast<ConstantVector>(C2);
822 if ((CP1 != NULL || isa<ConstantAggregateZero>(C1)) &&
823 (CP2 != NULL || isa<ConstantAggregateZero>(C2))) {
824 std::vector<Constant*> Res;
825 const Type* EltTy = VTy->getElementType();
826 const Constant *C1 = 0;
827 const Constant *C2 = 0;
828 switch (Opcode) {
829 default:
830 break;
831 case Instruction::Add:
832 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
833 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
834 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
835 Res.push_back(ConstantExpr::getAdd(const_cast<Constant*>(C1),
836 const_cast<Constant*>(C2)));
838 return ConstantVector::get(Res);
839 case Instruction::FAdd:
840 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
841 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
842 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
843 Res.push_back(ConstantExpr::getFAdd(const_cast<Constant*>(C1),
844 const_cast<Constant*>(C2)));
846 return ConstantVector::get(Res);
847 case Instruction::Sub:
848 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
849 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
850 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
851 Res.push_back(ConstantExpr::getSub(const_cast<Constant*>(C1),
852 const_cast<Constant*>(C2)));
854 return ConstantVector::get(Res);
855 case Instruction::FSub:
856 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
857 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
858 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
859 Res.push_back(ConstantExpr::getFSub(const_cast<Constant*>(C1),
860 const_cast<Constant*>(C2)));
862 return ConstantVector::get(Res);
863 case Instruction::Mul:
864 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
865 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
866 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
867 Res.push_back(ConstantExpr::getMul(const_cast<Constant*>(C1),
868 const_cast<Constant*>(C2)));
870 return ConstantVector::get(Res);
871 case Instruction::FMul:
872 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
873 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
874 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
875 Res.push_back(ConstantExpr::getFMul(const_cast<Constant*>(C1),
876 const_cast<Constant*>(C2)));
878 return ConstantVector::get(Res);
879 case Instruction::UDiv:
880 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
881 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
882 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
883 Res.push_back(ConstantExpr::getUDiv(const_cast<Constant*>(C1),
884 const_cast<Constant*>(C2)));
886 return ConstantVector::get(Res);
887 case Instruction::SDiv:
888 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
889 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
890 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
891 Res.push_back(ConstantExpr::getSDiv(const_cast<Constant*>(C1),
892 const_cast<Constant*>(C2)));
894 return ConstantVector::get(Res);
895 case Instruction::FDiv:
896 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
897 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
898 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
899 Res.push_back(ConstantExpr::getFDiv(const_cast<Constant*>(C1),
900 const_cast<Constant*>(C2)));
902 return ConstantVector::get(Res);
903 case Instruction::URem:
904 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
905 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
906 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
907 Res.push_back(ConstantExpr::getURem(const_cast<Constant*>(C1),
908 const_cast<Constant*>(C2)));
910 return ConstantVector::get(Res);
911 case Instruction::SRem:
912 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
913 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
914 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
915 Res.push_back(ConstantExpr::getSRem(const_cast<Constant*>(C1),
916 const_cast<Constant*>(C2)));
918 return ConstantVector::get(Res);
919 case Instruction::FRem:
920 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
921 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
922 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
923 Res.push_back(ConstantExpr::getFRem(const_cast<Constant*>(C1),
924 const_cast<Constant*>(C2)));
926 return ConstantVector::get(Res);
927 case Instruction::And:
928 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
929 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
930 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
931 Res.push_back(ConstantExpr::getAnd(const_cast<Constant*>(C1),
932 const_cast<Constant*>(C2)));
934 return ConstantVector::get(Res);
935 case Instruction::Or:
936 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
937 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
938 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
939 Res.push_back(ConstantExpr::getOr(const_cast<Constant*>(C1),
940 const_cast<Constant*>(C2)));
942 return ConstantVector::get(Res);
943 case Instruction::Xor:
944 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
945 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
946 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
947 Res.push_back(ConstantExpr::getXor(const_cast<Constant*>(C1),
948 const_cast<Constant*>(C2)));
950 return ConstantVector::get(Res);
951 case Instruction::LShr:
952 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
953 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
954 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
955 Res.push_back(ConstantExpr::getLShr(const_cast<Constant*>(C1),
956 const_cast<Constant*>(C2)));
958 return ConstantVector::get(Res);
959 case Instruction::AShr:
960 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
961 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
962 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
963 Res.push_back(ConstantExpr::getAShr(const_cast<Constant*>(C1),
964 const_cast<Constant*>(C2)));
966 return ConstantVector::get(Res);
967 case Instruction::Shl:
968 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) {
969 C1 = CP1 ? CP1->getOperand(i) : Constant::getNullValue(EltTy);
970 C2 = CP2 ? CP2->getOperand(i) : Constant::getNullValue(EltTy);
971 Res.push_back(ConstantExpr::getShl(const_cast<Constant*>(C1),
972 const_cast<Constant*>(C2)));
974 return ConstantVector::get(Res);
979 if (isa<ConstantExpr>(C1)) {
980 // There are many possible foldings we could do here. We should probably
981 // at least fold add of a pointer with an integer into the appropriate
982 // getelementptr. This will improve alias analysis a bit.
983 } else if (isa<ConstantExpr>(C2)) {
984 // If C2 is a constant expr and C1 isn't, flop them around and fold the
985 // other way if possible.
986 switch (Opcode) {
987 case Instruction::Add:
988 case Instruction::FAdd:
989 case Instruction::Mul:
990 case Instruction::FMul:
991 case Instruction::And:
992 case Instruction::Or:
993 case Instruction::Xor:
994 // No change of opcode required.
995 return ConstantFoldBinaryInstruction(Context, Opcode, C2, C1);
997 case Instruction::Shl:
998 case Instruction::LShr:
999 case Instruction::AShr:
1000 case Instruction::Sub:
1001 case Instruction::FSub:
1002 case Instruction::SDiv:
1003 case Instruction::UDiv:
1004 case Instruction::FDiv:
1005 case Instruction::URem:
1006 case Instruction::SRem:
1007 case Instruction::FRem:
1008 default: // These instructions cannot be flopped around.
1009 break;
1013 // We don't know how to fold this.
1014 return 0;
1017 /// isZeroSizedType - This type is zero sized if its an array or structure of
1018 /// zero sized types. The only leaf zero sized type is an empty structure.
1019 static bool isMaybeZeroSizedType(const Type *Ty) {
1020 if (isa<OpaqueType>(Ty)) return true; // Can't say.
1021 if (const StructType *STy = dyn_cast<StructType>(Ty)) {
1023 // If all of elements have zero size, this does too.
1024 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i)
1025 if (!isMaybeZeroSizedType(STy->getElementType(i))) return false;
1026 return true;
1028 } else if (const ArrayType *ATy = dyn_cast<ArrayType>(Ty)) {
1029 return isMaybeZeroSizedType(ATy->getElementType());
1031 return false;
1034 /// IdxCompare - Compare the two constants as though they were getelementptr
1035 /// indices. This allows coersion of the types to be the same thing.
1037 /// If the two constants are the "same" (after coersion), return 0. If the
1038 /// first is less than the second, return -1, if the second is less than the
1039 /// first, return 1. If the constants are not integral, return -2.
1041 static int IdxCompare(LLVMContext &Context, Constant *C1, Constant *C2,
1042 const Type *ElTy) {
1043 if (C1 == C2) return 0;
1045 // Ok, we found a different index. If they are not ConstantInt, we can't do
1046 // anything with them.
1047 if (!isa<ConstantInt>(C1) || !isa<ConstantInt>(C2))
1048 return -2; // don't know!
1050 // Ok, we have two differing integer indices. Sign extend them to be the same
1051 // type. Long is always big enough, so we use it.
1052 if (C1->getType() != Type::getInt64Ty(Context))
1053 C1 = ConstantExpr::getSExt(C1, Type::getInt64Ty(Context));
1055 if (C2->getType() != Type::getInt64Ty(Context))
1056 C2 = ConstantExpr::getSExt(C2, Type::getInt64Ty(Context));
1058 if (C1 == C2) return 0; // They are equal
1060 // If the type being indexed over is really just a zero sized type, there is
1061 // no pointer difference being made here.
1062 if (isMaybeZeroSizedType(ElTy))
1063 return -2; // dunno.
1065 // If they are really different, now that they are the same type, then we
1066 // found a difference!
1067 if (cast<ConstantInt>(C1)->getSExtValue() <
1068 cast<ConstantInt>(C2)->getSExtValue())
1069 return -1;
1070 else
1071 return 1;
1074 /// evaluateFCmpRelation - This function determines if there is anything we can
1075 /// decide about the two constants provided. This doesn't need to handle simple
1076 /// things like ConstantFP comparisons, but should instead handle ConstantExprs.
1077 /// If we can determine that the two constants have a particular relation to
1078 /// each other, we should return the corresponding FCmpInst predicate,
1079 /// otherwise return FCmpInst::BAD_FCMP_PREDICATE. This is used below in
1080 /// ConstantFoldCompareInstruction.
1082 /// To simplify this code we canonicalize the relation so that the first
1083 /// operand is always the most "complex" of the two. We consider ConstantFP
1084 /// to be the simplest, and ConstantExprs to be the most complex.
1085 static FCmpInst::Predicate evaluateFCmpRelation(LLVMContext &Context,
1086 const Constant *V1,
1087 const Constant *V2) {
1088 assert(V1->getType() == V2->getType() &&
1089 "Cannot compare values of different types!");
1091 // No compile-time operations on this type yet.
1092 if (V1->getType() == Type::getPPC_FP128Ty(Context))
1093 return FCmpInst::BAD_FCMP_PREDICATE;
1095 // Handle degenerate case quickly
1096 if (V1 == V2) return FCmpInst::FCMP_OEQ;
1098 if (!isa<ConstantExpr>(V1)) {
1099 if (!isa<ConstantExpr>(V2)) {
1100 // We distilled thisUse the standard constant folder for a few cases
1101 ConstantInt *R = 0;
1102 Constant *C1 = const_cast<Constant*>(V1);
1103 Constant *C2 = const_cast<Constant*>(V2);
1104 R = dyn_cast<ConstantInt>(
1105 ConstantExpr::getFCmp(FCmpInst::FCMP_OEQ, C1, C2));
1106 if (R && !R->isZero())
1107 return FCmpInst::FCMP_OEQ;
1108 R = dyn_cast<ConstantInt>(
1109 ConstantExpr::getFCmp(FCmpInst::FCMP_OLT, C1, C2));
1110 if (R && !R->isZero())
1111 return FCmpInst::FCMP_OLT;
1112 R = dyn_cast<ConstantInt>(
1113 ConstantExpr::getFCmp(FCmpInst::FCMP_OGT, C1, C2));
1114 if (R && !R->isZero())
1115 return FCmpInst::FCMP_OGT;
1117 // Nothing more we can do
1118 return FCmpInst::BAD_FCMP_PREDICATE;
1121 // If the first operand is simple and second is ConstantExpr, swap operands.
1122 FCmpInst::Predicate SwappedRelation = evaluateFCmpRelation(Context, V2, V1);
1123 if (SwappedRelation != FCmpInst::BAD_FCMP_PREDICATE)
1124 return FCmpInst::getSwappedPredicate(SwappedRelation);
1125 } else {
1126 // Ok, the LHS is known to be a constantexpr. The RHS can be any of a
1127 // constantexpr or a simple constant.
1128 const ConstantExpr *CE1 = cast<ConstantExpr>(V1);
1129 switch (CE1->getOpcode()) {
1130 case Instruction::FPTrunc:
1131 case Instruction::FPExt:
1132 case Instruction::UIToFP:
1133 case Instruction::SIToFP:
1134 // We might be able to do something with these but we don't right now.
1135 break;
1136 default:
1137 break;
1140 // There are MANY other foldings that we could perform here. They will
1141 // probably be added on demand, as they seem needed.
1142 return FCmpInst::BAD_FCMP_PREDICATE;
1145 /// evaluateICmpRelation - This function determines if there is anything we can
1146 /// decide about the two constants provided. This doesn't need to handle simple
1147 /// things like integer comparisons, but should instead handle ConstantExprs
1148 /// and GlobalValues. If we can determine that the two constants have a
1149 /// particular relation to each other, we should return the corresponding ICmp
1150 /// predicate, otherwise return ICmpInst::BAD_ICMP_PREDICATE.
1152 /// To simplify this code we canonicalize the relation so that the first
1153 /// operand is always the most "complex" of the two. We consider simple
1154 /// constants (like ConstantInt) to be the simplest, followed by
1155 /// GlobalValues, followed by ConstantExpr's (the most complex).
1157 static ICmpInst::Predicate evaluateICmpRelation(LLVMContext &Context,
1158 const Constant *V1,
1159 const Constant *V2,
1160 bool isSigned) {
1161 assert(V1->getType() == V2->getType() &&
1162 "Cannot compare different types of values!");
1163 if (V1 == V2) return ICmpInst::ICMP_EQ;
1165 if (!isa<ConstantExpr>(V1) && !isa<GlobalValue>(V1)) {
1166 if (!isa<GlobalValue>(V2) && !isa<ConstantExpr>(V2)) {
1167 // We distilled this down to a simple case, use the standard constant
1168 // folder.
1169 ConstantInt *R = 0;
1170 Constant *C1 = const_cast<Constant*>(V1);
1171 Constant *C2 = const_cast<Constant*>(V2);
1172 ICmpInst::Predicate pred = ICmpInst::ICMP_EQ;
1173 R = dyn_cast<ConstantInt>(ConstantExpr::getICmp(pred, C1, C2));
1174 if (R && !R->isZero())
1175 return pred;
1176 pred = isSigned ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT;
1177 R = dyn_cast<ConstantInt>(ConstantExpr::getICmp(pred, C1, C2));
1178 if (R && !R->isZero())
1179 return pred;
1180 pred = isSigned ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT;
1181 R = dyn_cast<ConstantInt>(ConstantExpr::getICmp(pred, C1, C2));
1182 if (R && !R->isZero())
1183 return pred;
1185 // If we couldn't figure it out, bail.
1186 return ICmpInst::BAD_ICMP_PREDICATE;
1189 // If the first operand is simple, swap operands.
1190 ICmpInst::Predicate SwappedRelation =
1191 evaluateICmpRelation(Context, V2, V1, isSigned);
1192 if (SwappedRelation != ICmpInst::BAD_ICMP_PREDICATE)
1193 return ICmpInst::getSwappedPredicate(SwappedRelation);
1195 } else if (const GlobalValue *CPR1 = dyn_cast<GlobalValue>(V1)) {
1196 if (isa<ConstantExpr>(V2)) { // Swap as necessary.
1197 ICmpInst::Predicate SwappedRelation =
1198 evaluateICmpRelation(Context, V2, V1, isSigned);
1199 if (SwappedRelation != ICmpInst::BAD_ICMP_PREDICATE)
1200 return ICmpInst::getSwappedPredicate(SwappedRelation);
1201 else
1202 return ICmpInst::BAD_ICMP_PREDICATE;
1205 // Now we know that the RHS is a GlobalValue or simple constant,
1206 // which (since the types must match) means that it's a ConstantPointerNull.
1207 if (const GlobalValue *CPR2 = dyn_cast<GlobalValue>(V2)) {
1208 // Don't try to decide equality of aliases.
1209 if (!isa<GlobalAlias>(CPR1) && !isa<GlobalAlias>(CPR2))
1210 if (!CPR1->hasExternalWeakLinkage() || !CPR2->hasExternalWeakLinkage())
1211 return ICmpInst::ICMP_NE;
1212 } else {
1213 assert(isa<ConstantPointerNull>(V2) && "Canonicalization guarantee!");
1214 // GlobalVals can never be null. Don't try to evaluate aliases.
1215 if (!CPR1->hasExternalWeakLinkage() && !isa<GlobalAlias>(CPR1))
1216 return ICmpInst::ICMP_NE;
1218 } else {
1219 // Ok, the LHS is known to be a constantexpr. The RHS can be any of a
1220 // constantexpr, a CPR, or a simple constant.
1221 const ConstantExpr *CE1 = cast<ConstantExpr>(V1);
1222 const Constant *CE1Op0 = CE1->getOperand(0);
1224 switch (CE1->getOpcode()) {
1225 case Instruction::Trunc:
1226 case Instruction::FPTrunc:
1227 case Instruction::FPExt:
1228 case Instruction::FPToUI:
1229 case Instruction::FPToSI:
1230 break; // We can't evaluate floating point casts or truncations.
1232 case Instruction::UIToFP:
1233 case Instruction::SIToFP:
1234 case Instruction::BitCast:
1235 case Instruction::ZExt:
1236 case Instruction::SExt:
1237 // If the cast is not actually changing bits, and the second operand is a
1238 // null pointer, do the comparison with the pre-casted value.
1239 if (V2->isNullValue() &&
1240 (isa<PointerType>(CE1->getType()) || CE1->getType()->isInteger())) {
1241 bool sgnd = isSigned;
1242 if (CE1->getOpcode() == Instruction::ZExt) isSigned = false;
1243 if (CE1->getOpcode() == Instruction::SExt) isSigned = true;
1244 return evaluateICmpRelation(Context, CE1Op0,
1245 Constant::getNullValue(CE1Op0->getType()),
1246 sgnd);
1249 // If the dest type is a pointer type, and the RHS is a constantexpr cast
1250 // from the same type as the src of the LHS, evaluate the inputs. This is
1251 // important for things like "icmp eq (cast 4 to int*), (cast 5 to int*)",
1252 // which happens a lot in compilers with tagged integers.
1253 if (const ConstantExpr *CE2 = dyn_cast<ConstantExpr>(V2))
1254 if (CE2->isCast() && isa<PointerType>(CE1->getType()) &&
1255 CE1->getOperand(0)->getType() == CE2->getOperand(0)->getType() &&
1256 CE1->getOperand(0)->getType()->isInteger()) {
1257 bool sgnd = isSigned;
1258 if (CE1->getOpcode() == Instruction::ZExt) isSigned = false;
1259 if (CE1->getOpcode() == Instruction::SExt) isSigned = true;
1260 return evaluateICmpRelation(Context, CE1->getOperand(0),
1261 CE2->getOperand(0), sgnd);
1263 break;
1265 case Instruction::GetElementPtr:
1266 // Ok, since this is a getelementptr, we know that the constant has a
1267 // pointer type. Check the various cases.
1268 if (isa<ConstantPointerNull>(V2)) {
1269 // If we are comparing a GEP to a null pointer, check to see if the base
1270 // of the GEP equals the null pointer.
1271 if (const GlobalValue *GV = dyn_cast<GlobalValue>(CE1Op0)) {
1272 if (GV->hasExternalWeakLinkage())
1273 // Weak linkage GVals could be zero or not. We're comparing that
1274 // to null pointer so its greater-or-equal
1275 return isSigned ? ICmpInst::ICMP_SGE : ICmpInst::ICMP_UGE;
1276 else
1277 // If its not weak linkage, the GVal must have a non-zero address
1278 // so the result is greater-than
1279 return isSigned ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT;
1280 } else if (isa<ConstantPointerNull>(CE1Op0)) {
1281 // If we are indexing from a null pointer, check to see if we have any
1282 // non-zero indices.
1283 for (unsigned i = 1, e = CE1->getNumOperands(); i != e; ++i)
1284 if (!CE1->getOperand(i)->isNullValue())
1285 // Offsetting from null, must not be equal.
1286 return isSigned ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT;
1287 // Only zero indexes from null, must still be zero.
1288 return ICmpInst::ICMP_EQ;
1290 // Otherwise, we can't really say if the first operand is null or not.
1291 } else if (const GlobalValue *CPR2 = dyn_cast<GlobalValue>(V2)) {
1292 if (isa<ConstantPointerNull>(CE1Op0)) {
1293 if (CPR2->hasExternalWeakLinkage())
1294 // Weak linkage GVals could be zero or not. We're comparing it to
1295 // a null pointer, so its less-or-equal
1296 return isSigned ? ICmpInst::ICMP_SLE : ICmpInst::ICMP_ULE;
1297 else
1298 // If its not weak linkage, the GVal must have a non-zero address
1299 // so the result is less-than
1300 return isSigned ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT;
1301 } else if (const GlobalValue *CPR1 = dyn_cast<GlobalValue>(CE1Op0)) {
1302 if (CPR1 == CPR2) {
1303 // If this is a getelementptr of the same global, then it must be
1304 // different. Because the types must match, the getelementptr could
1305 // only have at most one index, and because we fold getelementptr's
1306 // with a single zero index, it must be nonzero.
1307 assert(CE1->getNumOperands() == 2 &&
1308 !CE1->getOperand(1)->isNullValue() &&
1309 "Suprising getelementptr!");
1310 return isSigned ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT;
1311 } else {
1312 // If they are different globals, we don't know what the value is,
1313 // but they can't be equal.
1314 return ICmpInst::ICMP_NE;
1317 } else {
1318 const ConstantExpr *CE2 = cast<ConstantExpr>(V2);
1319 const Constant *CE2Op0 = CE2->getOperand(0);
1321 // There are MANY other foldings that we could perform here. They will
1322 // probably be added on demand, as they seem needed.
1323 switch (CE2->getOpcode()) {
1324 default: break;
1325 case Instruction::GetElementPtr:
1326 // By far the most common case to handle is when the base pointers are
1327 // obviously to the same or different globals.
1328 if (isa<GlobalValue>(CE1Op0) && isa<GlobalValue>(CE2Op0)) {
1329 if (CE1Op0 != CE2Op0) // Don't know relative ordering, but not equal
1330 return ICmpInst::ICMP_NE;
1331 // Ok, we know that both getelementptr instructions are based on the
1332 // same global. From this, we can precisely determine the relative
1333 // ordering of the resultant pointers.
1334 unsigned i = 1;
1336 // The logic below assumes that the result of the comparison
1337 // can be determined by finding the first index that differs.
1338 // This doesn't work if there is over-indexing in any
1339 // subsequent indices, so check for that case first.
1340 if (!CE1->isGEPWithNoNotionalOverIndexing() ||
1341 !CE2->isGEPWithNoNotionalOverIndexing())
1342 return ICmpInst::BAD_ICMP_PREDICATE; // Might be equal.
1344 // Compare all of the operands the GEP's have in common.
1345 gep_type_iterator GTI = gep_type_begin(CE1);
1346 for (;i != CE1->getNumOperands() && i != CE2->getNumOperands();
1347 ++i, ++GTI)
1348 switch (IdxCompare(Context, CE1->getOperand(i),
1349 CE2->getOperand(i), GTI.getIndexedType())) {
1350 case -1: return isSigned ? ICmpInst::ICMP_SLT:ICmpInst::ICMP_ULT;
1351 case 1: return isSigned ? ICmpInst::ICMP_SGT:ICmpInst::ICMP_UGT;
1352 case -2: return ICmpInst::BAD_ICMP_PREDICATE;
1355 // Ok, we ran out of things they have in common. If any leftovers
1356 // are non-zero then we have a difference, otherwise we are equal.
1357 for (; i < CE1->getNumOperands(); ++i)
1358 if (!CE1->getOperand(i)->isNullValue()) {
1359 if (isa<ConstantInt>(CE1->getOperand(i)))
1360 return isSigned ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT;
1361 else
1362 return ICmpInst::BAD_ICMP_PREDICATE; // Might be equal.
1365 for (; i < CE2->getNumOperands(); ++i)
1366 if (!CE2->getOperand(i)->isNullValue()) {
1367 if (isa<ConstantInt>(CE2->getOperand(i)))
1368 return isSigned ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT;
1369 else
1370 return ICmpInst::BAD_ICMP_PREDICATE; // Might be equal.
1372 return ICmpInst::ICMP_EQ;
1376 default:
1377 break;
1381 return ICmpInst::BAD_ICMP_PREDICATE;
1384 Constant *llvm::ConstantFoldCompareInstruction(LLVMContext &Context,
1385 unsigned short pred,
1386 const Constant *C1,
1387 const Constant *C2) {
1388 const Type *ResultTy;
1389 if (const VectorType *VT = dyn_cast<VectorType>(C1->getType()))
1390 ResultTy = VectorType::get(Type::getInt1Ty(Context), VT->getNumElements());
1391 else
1392 ResultTy = Type::getInt1Ty(Context);
1394 // Fold FCMP_FALSE/FCMP_TRUE unconditionally.
1395 if (pred == FCmpInst::FCMP_FALSE)
1396 return Constant::getNullValue(ResultTy);
1398 if (pred == FCmpInst::FCMP_TRUE)
1399 return Constant::getAllOnesValue(ResultTy);
1401 // Handle some degenerate cases first
1402 if (isa<UndefValue>(C1) || isa<UndefValue>(C2))
1403 return UndefValue::get(ResultTy);
1405 // No compile-time operations on this type yet.
1406 if (C1->getType() == Type::getPPC_FP128Ty(Context))
1407 return 0;
1409 // icmp eq/ne(null,GV) -> false/true
1410 if (C1->isNullValue()) {
1411 if (const GlobalValue *GV = dyn_cast<GlobalValue>(C2))
1412 // Don't try to evaluate aliases. External weak GV can be null.
1413 if (!isa<GlobalAlias>(GV) && !GV->hasExternalWeakLinkage()) {
1414 if (pred == ICmpInst::ICMP_EQ)
1415 return ConstantInt::getFalse(Context);
1416 else if (pred == ICmpInst::ICMP_NE)
1417 return ConstantInt::getTrue(Context);
1419 // icmp eq/ne(GV,null) -> false/true
1420 } else if (C2->isNullValue()) {
1421 if (const GlobalValue *GV = dyn_cast<GlobalValue>(C1))
1422 // Don't try to evaluate aliases. External weak GV can be null.
1423 if (!isa<GlobalAlias>(GV) && !GV->hasExternalWeakLinkage()) {
1424 if (pred == ICmpInst::ICMP_EQ)
1425 return ConstantInt::getFalse(Context);
1426 else if (pred == ICmpInst::ICMP_NE)
1427 return ConstantInt::getTrue(Context);
1431 if (isa<ConstantInt>(C1) && isa<ConstantInt>(C2)) {
1432 APInt V1 = cast<ConstantInt>(C1)->getValue();
1433 APInt V2 = cast<ConstantInt>(C2)->getValue();
1434 switch (pred) {
1435 default: llvm_unreachable("Invalid ICmp Predicate"); return 0;
1436 case ICmpInst::ICMP_EQ:
1437 return ConstantInt::get(Type::getInt1Ty(Context), V1 == V2);
1438 case ICmpInst::ICMP_NE:
1439 return ConstantInt::get(Type::getInt1Ty(Context), V1 != V2);
1440 case ICmpInst::ICMP_SLT:
1441 return ConstantInt::get(Type::getInt1Ty(Context), V1.slt(V2));
1442 case ICmpInst::ICMP_SGT:
1443 return ConstantInt::get(Type::getInt1Ty(Context), V1.sgt(V2));
1444 case ICmpInst::ICMP_SLE:
1445 return ConstantInt::get(Type::getInt1Ty(Context), V1.sle(V2));
1446 case ICmpInst::ICMP_SGE:
1447 return ConstantInt::get(Type::getInt1Ty(Context), V1.sge(V2));
1448 case ICmpInst::ICMP_ULT:
1449 return ConstantInt::get(Type::getInt1Ty(Context), V1.ult(V2));
1450 case ICmpInst::ICMP_UGT:
1451 return ConstantInt::get(Type::getInt1Ty(Context), V1.ugt(V2));
1452 case ICmpInst::ICMP_ULE:
1453 return ConstantInt::get(Type::getInt1Ty(Context), V1.ule(V2));
1454 case ICmpInst::ICMP_UGE:
1455 return ConstantInt::get(Type::getInt1Ty(Context), V1.uge(V2));
1457 } else if (isa<ConstantFP>(C1) && isa<ConstantFP>(C2)) {
1458 APFloat C1V = cast<ConstantFP>(C1)->getValueAPF();
1459 APFloat C2V = cast<ConstantFP>(C2)->getValueAPF();
1460 APFloat::cmpResult R = C1V.compare(C2V);
1461 switch (pred) {
1462 default: llvm_unreachable("Invalid FCmp Predicate"); return 0;
1463 case FCmpInst::FCMP_FALSE: return ConstantInt::getFalse(Context);
1464 case FCmpInst::FCMP_TRUE: return ConstantInt::getTrue(Context);
1465 case FCmpInst::FCMP_UNO:
1466 return ConstantInt::get(Type::getInt1Ty(Context), R==APFloat::cmpUnordered);
1467 case FCmpInst::FCMP_ORD:
1468 return ConstantInt::get(Type::getInt1Ty(Context), R!=APFloat::cmpUnordered);
1469 case FCmpInst::FCMP_UEQ:
1470 return ConstantInt::get(Type::getInt1Ty(Context), R==APFloat::cmpUnordered ||
1471 R==APFloat::cmpEqual);
1472 case FCmpInst::FCMP_OEQ:
1473 return ConstantInt::get(Type::getInt1Ty(Context), R==APFloat::cmpEqual);
1474 case FCmpInst::FCMP_UNE:
1475 return ConstantInt::get(Type::getInt1Ty(Context), R!=APFloat::cmpEqual);
1476 case FCmpInst::FCMP_ONE:
1477 return ConstantInt::get(Type::getInt1Ty(Context), R==APFloat::cmpLessThan ||
1478 R==APFloat::cmpGreaterThan);
1479 case FCmpInst::FCMP_ULT:
1480 return ConstantInt::get(Type::getInt1Ty(Context), R==APFloat::cmpUnordered ||
1481 R==APFloat::cmpLessThan);
1482 case FCmpInst::FCMP_OLT:
1483 return ConstantInt::get(Type::getInt1Ty(Context), R==APFloat::cmpLessThan);
1484 case FCmpInst::FCMP_UGT:
1485 return ConstantInt::get(Type::getInt1Ty(Context), R==APFloat::cmpUnordered ||
1486 R==APFloat::cmpGreaterThan);
1487 case FCmpInst::FCMP_OGT:
1488 return ConstantInt::get(Type::getInt1Ty(Context), R==APFloat::cmpGreaterThan);
1489 case FCmpInst::FCMP_ULE:
1490 return ConstantInt::get(Type::getInt1Ty(Context), R!=APFloat::cmpGreaterThan);
1491 case FCmpInst::FCMP_OLE:
1492 return ConstantInt::get(Type::getInt1Ty(Context), R==APFloat::cmpLessThan ||
1493 R==APFloat::cmpEqual);
1494 case FCmpInst::FCMP_UGE:
1495 return ConstantInt::get(Type::getInt1Ty(Context), R!=APFloat::cmpLessThan);
1496 case FCmpInst::FCMP_OGE:
1497 return ConstantInt::get(Type::getInt1Ty(Context), R==APFloat::cmpGreaterThan ||
1498 R==APFloat::cmpEqual);
1500 } else if (isa<VectorType>(C1->getType())) {
1501 SmallVector<Constant*, 16> C1Elts, C2Elts;
1502 C1->getVectorElements(Context, C1Elts);
1503 C2->getVectorElements(Context, C2Elts);
1505 // If we can constant fold the comparison of each element, constant fold
1506 // the whole vector comparison.
1507 SmallVector<Constant*, 4> ResElts;
1508 for (unsigned i = 0, e = C1Elts.size(); i != e; ++i) {
1509 // Compare the elements, producing an i1 result or constant expr.
1510 ResElts.push_back(
1511 ConstantExpr::getCompare(pred, C1Elts[i], C2Elts[i]));
1513 return ConstantVector::get(&ResElts[0], ResElts.size());
1516 if (C1->getType()->isFloatingPoint()) {
1517 int Result = -1; // -1 = unknown, 0 = known false, 1 = known true.
1518 switch (evaluateFCmpRelation(Context, C1, C2)) {
1519 default: llvm_unreachable("Unknown relation!");
1520 case FCmpInst::FCMP_UNO:
1521 case FCmpInst::FCMP_ORD:
1522 case FCmpInst::FCMP_UEQ:
1523 case FCmpInst::FCMP_UNE:
1524 case FCmpInst::FCMP_ULT:
1525 case FCmpInst::FCMP_UGT:
1526 case FCmpInst::FCMP_ULE:
1527 case FCmpInst::FCMP_UGE:
1528 case FCmpInst::FCMP_TRUE:
1529 case FCmpInst::FCMP_FALSE:
1530 case FCmpInst::BAD_FCMP_PREDICATE:
1531 break; // Couldn't determine anything about these constants.
1532 case FCmpInst::FCMP_OEQ: // We know that C1 == C2
1533 Result = (pred == FCmpInst::FCMP_UEQ || pred == FCmpInst::FCMP_OEQ ||
1534 pred == FCmpInst::FCMP_ULE || pred == FCmpInst::FCMP_OLE ||
1535 pred == FCmpInst::FCMP_UGE || pred == FCmpInst::FCMP_OGE);
1536 break;
1537 case FCmpInst::FCMP_OLT: // We know that C1 < C2
1538 Result = (pred == FCmpInst::FCMP_UNE || pred == FCmpInst::FCMP_ONE ||
1539 pred == FCmpInst::FCMP_ULT || pred == FCmpInst::FCMP_OLT ||
1540 pred == FCmpInst::FCMP_ULE || pred == FCmpInst::FCMP_OLE);
1541 break;
1542 case FCmpInst::FCMP_OGT: // We know that C1 > C2
1543 Result = (pred == FCmpInst::FCMP_UNE || pred == FCmpInst::FCMP_ONE ||
1544 pred == FCmpInst::FCMP_UGT || pred == FCmpInst::FCMP_OGT ||
1545 pred == FCmpInst::FCMP_UGE || pred == FCmpInst::FCMP_OGE);
1546 break;
1547 case FCmpInst::FCMP_OLE: // We know that C1 <= C2
1548 // We can only partially decide this relation.
1549 if (pred == FCmpInst::FCMP_UGT || pred == FCmpInst::FCMP_OGT)
1550 Result = 0;
1551 else if (pred == FCmpInst::FCMP_ULT || pred == FCmpInst::FCMP_OLT)
1552 Result = 1;
1553 break;
1554 case FCmpInst::FCMP_OGE: // We known that C1 >= C2
1555 // We can only partially decide this relation.
1556 if (pred == FCmpInst::FCMP_ULT || pred == FCmpInst::FCMP_OLT)
1557 Result = 0;
1558 else if (pred == FCmpInst::FCMP_UGT || pred == FCmpInst::FCMP_OGT)
1559 Result = 1;
1560 break;
1561 case ICmpInst::ICMP_NE: // We know that C1 != C2
1562 // We can only partially decide this relation.
1563 if (pred == FCmpInst::FCMP_OEQ || pred == FCmpInst::FCMP_UEQ)
1564 Result = 0;
1565 else if (pred == FCmpInst::FCMP_ONE || pred == FCmpInst::FCMP_UNE)
1566 Result = 1;
1567 break;
1570 // If we evaluated the result, return it now.
1571 if (Result != -1)
1572 return ConstantInt::get(Type::getInt1Ty(Context), Result);
1574 } else {
1575 // Evaluate the relation between the two constants, per the predicate.
1576 int Result = -1; // -1 = unknown, 0 = known false, 1 = known true.
1577 switch (evaluateICmpRelation(Context, C1, C2, CmpInst::isSigned(pred))) {
1578 default: llvm_unreachable("Unknown relational!");
1579 case ICmpInst::BAD_ICMP_PREDICATE:
1580 break; // Couldn't determine anything about these constants.
1581 case ICmpInst::ICMP_EQ: // We know the constants are equal!
1582 // If we know the constants are equal, we can decide the result of this
1583 // computation precisely.
1584 Result = (pred == ICmpInst::ICMP_EQ ||
1585 pred == ICmpInst::ICMP_ULE ||
1586 pred == ICmpInst::ICMP_SLE ||
1587 pred == ICmpInst::ICMP_UGE ||
1588 pred == ICmpInst::ICMP_SGE);
1589 break;
1590 case ICmpInst::ICMP_ULT:
1591 // If we know that C1 < C2, we can decide the result of this computation
1592 // precisely.
1593 Result = (pred == ICmpInst::ICMP_ULT ||
1594 pred == ICmpInst::ICMP_NE ||
1595 pred == ICmpInst::ICMP_ULE);
1596 break;
1597 case ICmpInst::ICMP_SLT:
1598 // If we know that C1 < C2, we can decide the result of this computation
1599 // precisely.
1600 Result = (pred == ICmpInst::ICMP_SLT ||
1601 pred == ICmpInst::ICMP_NE ||
1602 pred == ICmpInst::ICMP_SLE);
1603 break;
1604 case ICmpInst::ICMP_UGT:
1605 // If we know that C1 > C2, we can decide the result of this computation
1606 // precisely.
1607 Result = (pred == ICmpInst::ICMP_UGT ||
1608 pred == ICmpInst::ICMP_NE ||
1609 pred == ICmpInst::ICMP_UGE);
1610 break;
1611 case ICmpInst::ICMP_SGT:
1612 // If we know that C1 > C2, we can decide the result of this computation
1613 // precisely.
1614 Result = (pred == ICmpInst::ICMP_SGT ||
1615 pred == ICmpInst::ICMP_NE ||
1616 pred == ICmpInst::ICMP_SGE);
1617 break;
1618 case ICmpInst::ICMP_ULE:
1619 // If we know that C1 <= C2, we can only partially decide this relation.
1620 if (pred == ICmpInst::ICMP_UGT) Result = 0;
1621 if (pred == ICmpInst::ICMP_ULT) Result = 1;
1622 break;
1623 case ICmpInst::ICMP_SLE:
1624 // If we know that C1 <= C2, we can only partially decide this relation.
1625 if (pred == ICmpInst::ICMP_SGT) Result = 0;
1626 if (pred == ICmpInst::ICMP_SLT) Result = 1;
1627 break;
1629 case ICmpInst::ICMP_UGE:
1630 // If we know that C1 >= C2, we can only partially decide this relation.
1631 if (pred == ICmpInst::ICMP_ULT) Result = 0;
1632 if (pred == ICmpInst::ICMP_UGT) Result = 1;
1633 break;
1634 case ICmpInst::ICMP_SGE:
1635 // If we know that C1 >= C2, we can only partially decide this relation.
1636 if (pred == ICmpInst::ICMP_SLT) Result = 0;
1637 if (pred == ICmpInst::ICMP_SGT) Result = 1;
1638 break;
1640 case ICmpInst::ICMP_NE:
1641 // If we know that C1 != C2, we can only partially decide this relation.
1642 if (pred == ICmpInst::ICMP_EQ) Result = 0;
1643 if (pred == ICmpInst::ICMP_NE) Result = 1;
1644 break;
1647 // If we evaluated the result, return it now.
1648 if (Result != -1)
1649 return ConstantInt::get(Type::getInt1Ty(Context), Result);
1651 if (!isa<ConstantExpr>(C1) && isa<ConstantExpr>(C2)) {
1652 // If C2 is a constant expr and C1 isn't, flip them around and fold the
1653 // other way if possible.
1654 switch (pred) {
1655 case ICmpInst::ICMP_EQ:
1656 case ICmpInst::ICMP_NE:
1657 // No change of predicate required.
1658 return ConstantFoldCompareInstruction(Context, pred, C2, C1);
1660 case ICmpInst::ICMP_ULT:
1661 case ICmpInst::ICMP_SLT:
1662 case ICmpInst::ICMP_UGT:
1663 case ICmpInst::ICMP_SGT:
1664 case ICmpInst::ICMP_ULE:
1665 case ICmpInst::ICMP_SLE:
1666 case ICmpInst::ICMP_UGE:
1667 case ICmpInst::ICMP_SGE:
1668 // Change the predicate as necessary to swap the operands.
1669 pred = ICmpInst::getSwappedPredicate((ICmpInst::Predicate)pred);
1670 return ConstantFoldCompareInstruction(Context, pred, C2, C1);
1672 default: // These predicates cannot be flopped around.
1673 break;
1677 return 0;
1680 /// isInBoundsIndices - Test whether the given sequence of *normalized* indices
1681 /// is "inbounds".
1682 static bool isInBoundsIndices(Constant *const *Idxs, size_t NumIdx) {
1683 // No indices means nothing that could be out of bounds.
1684 if (NumIdx == 0) return true;
1686 // If the first index is zero, it's in bounds.
1687 if (Idxs[0]->isNullValue()) return true;
1689 // If the first index is one and all the rest are zero, it's in bounds,
1690 // by the one-past-the-end rule.
1691 if (!cast<ConstantInt>(Idxs[0])->isOne())
1692 return false;
1693 for (unsigned i = 1, e = NumIdx; i != e; ++i)
1694 if (!Idxs[i]->isNullValue())
1695 return false;
1696 return true;
1699 Constant *llvm::ConstantFoldGetElementPtr(LLVMContext &Context,
1700 const Constant *C,
1701 bool inBounds,
1702 Constant* const *Idxs,
1703 unsigned NumIdx) {
1704 if (NumIdx == 0 ||
1705 (NumIdx == 1 && Idxs[0]->isNullValue()))
1706 return const_cast<Constant*>(C);
1708 if (isa<UndefValue>(C)) {
1709 const PointerType *Ptr = cast<PointerType>(C->getType());
1710 const Type *Ty = GetElementPtrInst::getIndexedType(Ptr,
1711 (Value **)Idxs,
1712 (Value **)Idxs+NumIdx);
1713 assert(Ty != 0 && "Invalid indices for GEP!");
1714 return UndefValue::get(PointerType::get(Ty, Ptr->getAddressSpace()));
1717 Constant *Idx0 = Idxs[0];
1718 if (C->isNullValue()) {
1719 bool isNull = true;
1720 for (unsigned i = 0, e = NumIdx; i != e; ++i)
1721 if (!Idxs[i]->isNullValue()) {
1722 isNull = false;
1723 break;
1725 if (isNull) {
1726 const PointerType *Ptr = cast<PointerType>(C->getType());
1727 const Type *Ty = GetElementPtrInst::getIndexedType(Ptr,
1728 (Value**)Idxs,
1729 (Value**)Idxs+NumIdx);
1730 assert(Ty != 0 && "Invalid indices for GEP!");
1731 return ConstantPointerNull::get(
1732 PointerType::get(Ty,Ptr->getAddressSpace()));
1736 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(const_cast<Constant*>(C))) {
1737 // Combine Indices - If the source pointer to this getelementptr instruction
1738 // is a getelementptr instruction, combine the indices of the two
1739 // getelementptr instructions into a single instruction.
1741 if (CE->getOpcode() == Instruction::GetElementPtr) {
1742 const Type *LastTy = 0;
1743 for (gep_type_iterator I = gep_type_begin(CE), E = gep_type_end(CE);
1744 I != E; ++I)
1745 LastTy = *I;
1747 if ((LastTy && isa<ArrayType>(LastTy)) || Idx0->isNullValue()) {
1748 SmallVector<Value*, 16> NewIndices;
1749 NewIndices.reserve(NumIdx + CE->getNumOperands());
1750 for (unsigned i = 1, e = CE->getNumOperands()-1; i != e; ++i)
1751 NewIndices.push_back(CE->getOperand(i));
1753 // Add the last index of the source with the first index of the new GEP.
1754 // Make sure to handle the case when they are actually different types.
1755 Constant *Combined = CE->getOperand(CE->getNumOperands()-1);
1756 // Otherwise it must be an array.
1757 if (!Idx0->isNullValue()) {
1758 const Type *IdxTy = Combined->getType();
1759 if (IdxTy != Idx0->getType()) {
1760 Constant *C1 =
1761 ConstantExpr::getSExtOrBitCast(Idx0, Type::getInt64Ty(Context));
1762 Constant *C2 = ConstantExpr::getSExtOrBitCast(Combined,
1763 Type::getInt64Ty(Context));
1764 Combined = ConstantExpr::get(Instruction::Add, C1, C2);
1765 } else {
1766 Combined =
1767 ConstantExpr::get(Instruction::Add, Idx0, Combined);
1771 NewIndices.push_back(Combined);
1772 NewIndices.insert(NewIndices.end(), Idxs+1, Idxs+NumIdx);
1773 return (inBounds && cast<GEPOperator>(CE)->isInBounds()) ?
1774 ConstantExpr::getInBoundsGetElementPtr(CE->getOperand(0),
1775 &NewIndices[0],
1776 NewIndices.size()) :
1777 ConstantExpr::getGetElementPtr(CE->getOperand(0),
1778 &NewIndices[0],
1779 NewIndices.size());
1783 // Implement folding of:
1784 // int* getelementptr ([2 x int]* cast ([3 x int]* %X to [2 x int]*),
1785 // long 0, long 0)
1786 // To: int* getelementptr ([3 x int]* %X, long 0, long 0)
1788 if (CE->isCast() && NumIdx > 1 && Idx0->isNullValue()) {
1789 if (const PointerType *SPT =
1790 dyn_cast<PointerType>(CE->getOperand(0)->getType()))
1791 if (const ArrayType *SAT = dyn_cast<ArrayType>(SPT->getElementType()))
1792 if (const ArrayType *CAT =
1793 dyn_cast<ArrayType>(cast<PointerType>(C->getType())->getElementType()))
1794 if (CAT->getElementType() == SAT->getElementType())
1795 return inBounds ?
1796 ConstantExpr::getInBoundsGetElementPtr(
1797 (Constant*)CE->getOperand(0), Idxs, NumIdx) :
1798 ConstantExpr::getGetElementPtr(
1799 (Constant*)CE->getOperand(0), Idxs, NumIdx);
1802 // Fold: getelementptr (i8* inttoptr (i64 1 to i8*), i32 -1)
1803 // Into: inttoptr (i64 0 to i8*)
1804 // This happens with pointers to member functions in C++.
1805 if (CE->getOpcode() == Instruction::IntToPtr && NumIdx == 1 &&
1806 isa<ConstantInt>(CE->getOperand(0)) && isa<ConstantInt>(Idxs[0]) &&
1807 cast<PointerType>(CE->getType())->getElementType() == Type::getInt8Ty(Context)) {
1808 Constant *Base = CE->getOperand(0);
1809 Constant *Offset = Idxs[0];
1811 // Convert the smaller integer to the larger type.
1812 if (Offset->getType()->getPrimitiveSizeInBits() <
1813 Base->getType()->getPrimitiveSizeInBits())
1814 Offset = ConstantExpr::getSExt(Offset, Base->getType());
1815 else if (Base->getType()->getPrimitiveSizeInBits() <
1816 Offset->getType()->getPrimitiveSizeInBits())
1817 Base = ConstantExpr::getZExt(Base, Offset->getType());
1819 Base = ConstantExpr::getAdd(Base, Offset);
1820 return ConstantExpr::getIntToPtr(Base, CE->getType());
1824 // Check to see if any array indices are not within the corresponding
1825 // notional array bounds. If so, try to determine if they can be factored
1826 // out into preceding dimensions.
1827 bool Unknown = false;
1828 SmallVector<Constant *, 8> NewIdxs;
1829 const Type *Ty = C->getType();
1830 const Type *Prev = 0;
1831 for (unsigned i = 0; i != NumIdx;
1832 Prev = Ty, Ty = cast<CompositeType>(Ty)->getTypeAtIndex(Idxs[i]), ++i) {
1833 if (ConstantInt *CI = dyn_cast<ConstantInt>(Idxs[i])) {
1834 if (const ArrayType *ATy = dyn_cast<ArrayType>(Ty))
1835 if (ATy->getNumElements() <= INT64_MAX &&
1836 ATy->getNumElements() != 0 &&
1837 CI->getSExtValue() >= (int64_t)ATy->getNumElements()) {
1838 if (isa<SequentialType>(Prev)) {
1839 // It's out of range, but we can factor it into the prior
1840 // dimension.
1841 NewIdxs.resize(NumIdx);
1842 ConstantInt *Factor = ConstantInt::get(CI->getType(),
1843 ATy->getNumElements());
1844 NewIdxs[i] = ConstantExpr::getSRem(CI, Factor);
1846 Constant *PrevIdx = Idxs[i-1];
1847 Constant *Div = ConstantExpr::getSDiv(CI, Factor);
1849 // Before adding, extend both operands to i64 to avoid
1850 // overflow trouble.
1851 if (PrevIdx->getType() != Type::getInt64Ty(Context))
1852 PrevIdx = ConstantExpr::getSExt(PrevIdx,
1853 Type::getInt64Ty(Context));
1854 if (Div->getType() != Type::getInt64Ty(Context))
1855 Div = ConstantExpr::getSExt(Div,
1856 Type::getInt64Ty(Context));
1858 NewIdxs[i-1] = ConstantExpr::getAdd(PrevIdx, Div);
1859 } else {
1860 // It's out of range, but the prior dimension is a struct
1861 // so we can't do anything about it.
1862 Unknown = true;
1865 } else {
1866 // We don't know if it's in range or not.
1867 Unknown = true;
1871 // If we did any factoring, start over with the adjusted indices.
1872 if (!NewIdxs.empty()) {
1873 for (unsigned i = 0; i != NumIdx; ++i)
1874 if (!NewIdxs[i]) NewIdxs[i] = Idxs[i];
1875 return inBounds ?
1876 ConstantExpr::getInBoundsGetElementPtr(const_cast<Constant*>(C),
1877 NewIdxs.data(), NewIdxs.size()) :
1878 ConstantExpr::getGetElementPtr(const_cast<Constant*>(C),
1879 NewIdxs.data(), NewIdxs.size());
1882 // If all indices are known integers and normalized, we can do a simple
1883 // check for the "inbounds" property.
1884 if (!Unknown && !inBounds &&
1885 isa<GlobalVariable>(C) && isInBoundsIndices(Idxs, NumIdx))
1886 return ConstantExpr::getInBoundsGetElementPtr(const_cast<Constant*>(C),
1887 Idxs, NumIdx);
1889 return 0;