1 //===- ValueTracking.cpp - Walk computations to compute properties --------===//
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 contains routines that help analyze properties that chains of
13 //===----------------------------------------------------------------------===//
15 #include "llvm/Analysis/ValueTracking.h"
16 #include "llvm/ADT/SmallPtrSet.h"
17 #include "llvm/Analysis/InstructionSimplify.h"
18 #include "llvm/Analysis/MemoryBuiltins.h"
19 #include "llvm/IR/Constants.h"
20 #include "llvm/IR/DataLayout.h"
21 #include "llvm/IR/GlobalAlias.h"
22 #include "llvm/IR/GlobalVariable.h"
23 #include "llvm/IR/Instructions.h"
24 #include "llvm/IR/IntrinsicInst.h"
25 #include "llvm/IR/LLVMContext.h"
26 #include "llvm/IR/Metadata.h"
27 #include "llvm/IR/Operator.h"
28 #include "llvm/Support/ConstantRange.h"
29 #include "llvm/Support/GetElementPtrTypeIterator.h"
30 #include "llvm/Support/MathExtras.h"
31 #include "llvm/Support/PatternMatch.h"
34 using namespace llvm::PatternMatch;
36 const unsigned MaxDepth = 6;
38 /// getBitWidth - Returns the bitwidth of the given scalar or pointer type (if
39 /// unknown returns 0). For vector types, returns the element type's bitwidth.
40 static unsigned getBitWidth(Type *Ty, const DataLayout *TD) {
41 if (unsigned BitWidth = Ty->getScalarSizeInBits())
44 return TD ? TD->getPointerTypeSizeInBits(Ty) : 0;
47 static void ComputeMaskedBitsAddSub(bool Add, Value *Op0, Value *Op1, bool NSW,
48 APInt &KnownZero, APInt &KnownOne,
49 APInt &KnownZero2, APInt &KnownOne2,
50 const DataLayout *TD, unsigned Depth) {
52 if (ConstantInt *CLHS = dyn_cast<ConstantInt>(Op0)) {
53 // We know that the top bits of C-X are clear if X contains less bits
54 // than C (i.e. no wrap-around can happen). For example, 20-X is
55 // positive if we can prove that X is >= 0 and < 16.
56 if (!CLHS->getValue().isNegative()) {
57 unsigned BitWidth = KnownZero.getBitWidth();
58 unsigned NLZ = (CLHS->getValue()+1).countLeadingZeros();
59 // NLZ can't be BitWidth with no sign bit
60 APInt MaskV = APInt::getHighBitsSet(BitWidth, NLZ+1);
61 llvm::ComputeMaskedBits(Op1, KnownZero2, KnownOne2, TD, Depth+1);
63 // If all of the MaskV bits are known to be zero, then we know the
64 // output top bits are zero, because we now know that the output is
66 if ((KnownZero2 & MaskV) == MaskV) {
67 unsigned NLZ2 = CLHS->getValue().countLeadingZeros();
68 // Top bits known zero.
69 KnownZero = APInt::getHighBitsSet(BitWidth, NLZ2);
75 unsigned BitWidth = KnownZero.getBitWidth();
77 // If one of the operands has trailing zeros, then the bits that the
78 // other operand has in those bit positions will be preserved in the
79 // result. For an add, this works with either operand. For a subtract,
80 // this only works if the known zeros are in the right operand.
81 APInt LHSKnownZero(BitWidth, 0), LHSKnownOne(BitWidth, 0);
82 llvm::ComputeMaskedBits(Op0, LHSKnownZero, LHSKnownOne, TD, Depth+1);
83 assert((LHSKnownZero & LHSKnownOne) == 0 &&
84 "Bits known to be one AND zero?");
85 unsigned LHSKnownZeroOut = LHSKnownZero.countTrailingOnes();
87 llvm::ComputeMaskedBits(Op1, KnownZero2, KnownOne2, TD, Depth+1);
88 assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
89 unsigned RHSKnownZeroOut = KnownZero2.countTrailingOnes();
91 // Determine which operand has more trailing zeros, and use that
92 // many bits from the other operand.
93 if (LHSKnownZeroOut > RHSKnownZeroOut) {
95 APInt Mask = APInt::getLowBitsSet(BitWidth, LHSKnownZeroOut);
96 KnownZero |= KnownZero2 & Mask;
97 KnownOne |= KnownOne2 & Mask;
99 // If the known zeros are in the left operand for a subtract,
100 // fall back to the minimum known zeros in both operands.
101 KnownZero |= APInt::getLowBitsSet(BitWidth,
102 std::min(LHSKnownZeroOut,
105 } else if (RHSKnownZeroOut >= LHSKnownZeroOut) {
106 APInt Mask = APInt::getLowBitsSet(BitWidth, RHSKnownZeroOut);
107 KnownZero |= LHSKnownZero & Mask;
108 KnownOne |= LHSKnownOne & Mask;
111 // Are we still trying to solve for the sign bit?
112 if (!KnownZero.isNegative() && !KnownOne.isNegative()) {
115 // Adding two positive numbers can't wrap into negative
116 if (LHSKnownZero.isNegative() && KnownZero2.isNegative())
117 KnownZero |= APInt::getSignBit(BitWidth);
118 // and adding two negative numbers can't wrap into positive.
119 else if (LHSKnownOne.isNegative() && KnownOne2.isNegative())
120 KnownOne |= APInt::getSignBit(BitWidth);
122 // Subtracting a negative number from a positive one can't wrap
123 if (LHSKnownZero.isNegative() && KnownOne2.isNegative())
124 KnownZero |= APInt::getSignBit(BitWidth);
125 // neither can subtracting a positive number from a negative one.
126 else if (LHSKnownOne.isNegative() && KnownZero2.isNegative())
127 KnownOne |= APInt::getSignBit(BitWidth);
133 static void ComputeMaskedBitsMul(Value *Op0, Value *Op1, bool NSW,
134 APInt &KnownZero, APInt &KnownOne,
135 APInt &KnownZero2, APInt &KnownOne2,
136 const DataLayout *TD, unsigned Depth) {
137 unsigned BitWidth = KnownZero.getBitWidth();
138 ComputeMaskedBits(Op1, KnownZero, KnownOne, TD, Depth+1);
139 ComputeMaskedBits(Op0, KnownZero2, KnownOne2, TD, Depth+1);
140 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
141 assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
143 bool isKnownNegative = false;
144 bool isKnownNonNegative = false;
145 // If the multiplication is known not to overflow, compute the sign bit.
148 // The product of a number with itself is non-negative.
149 isKnownNonNegative = true;
151 bool isKnownNonNegativeOp1 = KnownZero.isNegative();
152 bool isKnownNonNegativeOp0 = KnownZero2.isNegative();
153 bool isKnownNegativeOp1 = KnownOne.isNegative();
154 bool isKnownNegativeOp0 = KnownOne2.isNegative();
155 // The product of two numbers with the same sign is non-negative.
156 isKnownNonNegative = (isKnownNegativeOp1 && isKnownNegativeOp0) ||
157 (isKnownNonNegativeOp1 && isKnownNonNegativeOp0);
158 // The product of a negative number and a non-negative number is either
160 if (!isKnownNonNegative)
161 isKnownNegative = (isKnownNegativeOp1 && isKnownNonNegativeOp0 &&
162 isKnownNonZero(Op0, TD, Depth)) ||
163 (isKnownNegativeOp0 && isKnownNonNegativeOp1 &&
164 isKnownNonZero(Op1, TD, Depth));
168 // If low bits are zero in either operand, output low known-0 bits.
169 // Also compute a conserative estimate for high known-0 bits.
170 // More trickiness is possible, but this is sufficient for the
171 // interesting case of alignment computation.
172 KnownOne.clearAllBits();
173 unsigned TrailZ = KnownZero.countTrailingOnes() +
174 KnownZero2.countTrailingOnes();
175 unsigned LeadZ = std::max(KnownZero.countLeadingOnes() +
176 KnownZero2.countLeadingOnes(),
177 BitWidth) - BitWidth;
179 TrailZ = std::min(TrailZ, BitWidth);
180 LeadZ = std::min(LeadZ, BitWidth);
181 KnownZero = APInt::getLowBitsSet(BitWidth, TrailZ) |
182 APInt::getHighBitsSet(BitWidth, LeadZ);
184 // Only make use of no-wrap flags if we failed to compute the sign bit
185 // directly. This matters if the multiplication always overflows, in
186 // which case we prefer to follow the result of the direct computation,
187 // though as the program is invoking undefined behaviour we can choose
188 // whatever we like here.
189 if (isKnownNonNegative && !KnownOne.isNegative())
190 KnownZero.setBit(BitWidth - 1);
191 else if (isKnownNegative && !KnownZero.isNegative())
192 KnownOne.setBit(BitWidth - 1);
195 void llvm::computeMaskedBitsLoad(const MDNode &Ranges, APInt &KnownZero) {
196 unsigned BitWidth = KnownZero.getBitWidth();
197 unsigned NumRanges = Ranges.getNumOperands() / 2;
198 assert(NumRanges >= 1);
200 // Use the high end of the ranges to find leading zeros.
201 unsigned MinLeadingZeros = BitWidth;
202 for (unsigned i = 0; i < NumRanges; ++i) {
203 ConstantInt *Lower = cast<ConstantInt>(Ranges.getOperand(2*i + 0));
204 ConstantInt *Upper = cast<ConstantInt>(Ranges.getOperand(2*i + 1));
205 ConstantRange Range(Lower->getValue(), Upper->getValue());
206 if (Range.isWrappedSet())
207 MinLeadingZeros = 0; // -1 has no zeros
208 unsigned LeadingZeros = (Upper->getValue() - 1).countLeadingZeros();
209 MinLeadingZeros = std::min(LeadingZeros, MinLeadingZeros);
212 KnownZero = APInt::getHighBitsSet(BitWidth, MinLeadingZeros);
214 /// ComputeMaskedBits - Determine which of the bits are known to be either zero
215 /// or one and return them in the KnownZero/KnownOne bit sets.
217 /// NOTE: we cannot consider 'undef' to be "IsZero" here. The problem is that
218 /// we cannot optimize based on the assumption that it is zero without changing
219 /// it to be an explicit zero. If we don't change it to zero, other code could
220 /// optimized based on the contradictory assumption that it is non-zero.
221 /// Because instcombine aggressively folds operations with undef args anyway,
222 /// this won't lose us code quality.
224 /// This function is defined on values with integer type, values with pointer
225 /// type (but only if TD is non-null), and vectors of integers. In the case
226 /// where V is a vector, known zero, and known one values are the
227 /// same width as the vector element, and the bit is set only if it is true
228 /// for all of the elements in the vector.
229 void llvm::ComputeMaskedBits(Value *V, APInt &KnownZero, APInt &KnownOne,
230 const DataLayout *TD, unsigned Depth) {
231 assert(V && "No Value?");
232 assert(Depth <= MaxDepth && "Limit Search Depth");
233 unsigned BitWidth = KnownZero.getBitWidth();
235 assert((V->getType()->isIntOrIntVectorTy() ||
236 V->getType()->getScalarType()->isPointerTy()) &&
237 "Not integer or pointer type!");
239 TD->getTypeSizeInBits(V->getType()->getScalarType()) == BitWidth) &&
240 (!V->getType()->isIntOrIntVectorTy() ||
241 V->getType()->getScalarSizeInBits() == BitWidth) &&
242 KnownZero.getBitWidth() == BitWidth &&
243 KnownOne.getBitWidth() == BitWidth &&
244 "V, Mask, KnownOne and KnownZero should have same BitWidth");
246 if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
247 // We know all of the bits for a constant!
248 KnownOne = CI->getValue();
249 KnownZero = ~KnownOne;
252 // Null and aggregate-zero are all-zeros.
253 if (isa<ConstantPointerNull>(V) ||
254 isa<ConstantAggregateZero>(V)) {
255 KnownOne.clearAllBits();
256 KnownZero = APInt::getAllOnesValue(BitWidth);
259 // Handle a constant vector by taking the intersection of the known bits of
260 // each element. There is no real need to handle ConstantVector here, because
261 // we don't handle undef in any particularly useful way.
262 if (ConstantDataSequential *CDS = dyn_cast<ConstantDataSequential>(V)) {
263 // We know that CDS must be a vector of integers. Take the intersection of
265 KnownZero.setAllBits(); KnownOne.setAllBits();
266 APInt Elt(KnownZero.getBitWidth(), 0);
267 for (unsigned i = 0, e = CDS->getNumElements(); i != e; ++i) {
268 Elt = CDS->getElementAsInteger(i);
275 // The address of an aligned GlobalValue has trailing zeros.
276 if (GlobalValue *GV = dyn_cast<GlobalValue>(V)) {
277 unsigned Align = GV->getAlignment();
278 if (Align == 0 && TD) {
279 if (GlobalVariable *GVar = dyn_cast<GlobalVariable>(GV)) {
280 Type *ObjectType = GVar->getType()->getElementType();
281 if (ObjectType->isSized()) {
282 // If the object is defined in the current Module, we'll be giving
283 // it the preferred alignment. Otherwise, we have to assume that it
284 // may only have the minimum ABI alignment.
285 if (!GVar->isDeclaration() && !GVar->isWeakForLinker())
286 Align = TD->getPreferredAlignment(GVar);
288 Align = TD->getABITypeAlignment(ObjectType);
293 KnownZero = APInt::getLowBitsSet(BitWidth,
294 countTrailingZeros(Align));
296 KnownZero.clearAllBits();
297 KnownOne.clearAllBits();
300 // A weak GlobalAlias is totally unknown. A non-weak GlobalAlias has
301 // the bits of its aliasee.
302 if (GlobalAlias *GA = dyn_cast<GlobalAlias>(V)) {
303 if (GA->mayBeOverridden()) {
304 KnownZero.clearAllBits(); KnownOne.clearAllBits();
306 ComputeMaskedBits(GA->getAliasee(), KnownZero, KnownOne, TD, Depth+1);
311 if (Argument *A = dyn_cast<Argument>(V)) {
314 if (A->hasByValAttr()) {
315 // Get alignment information off byval arguments if specified in the IR.
316 Align = A->getParamAlignment();
317 } else if (TD && A->hasStructRetAttr()) {
318 // An sret parameter has at least the ABI alignment of the return type.
319 Type *EltTy = cast<PointerType>(A->getType())->getElementType();
320 if (EltTy->isSized())
321 Align = TD->getABITypeAlignment(EltTy);
325 KnownZero = APInt::getLowBitsSet(BitWidth, countTrailingZeros(Align));
329 // Start out not knowing anything.
330 KnownZero.clearAllBits(); KnownOne.clearAllBits();
332 if (Depth == MaxDepth)
333 return; // Limit search depth.
335 Operator *I = dyn_cast<Operator>(V);
338 APInt KnownZero2(KnownZero), KnownOne2(KnownOne);
339 switch (I->getOpcode()) {
341 case Instruction::Load:
342 if (MDNode *MD = cast<LoadInst>(I)->getMetadata(LLVMContext::MD_range))
343 computeMaskedBitsLoad(*MD, KnownZero);
345 case Instruction::And: {
346 // If either the LHS or the RHS are Zero, the result is zero.
347 ComputeMaskedBits(I->getOperand(1), KnownZero, KnownOne, TD, Depth+1);
348 ComputeMaskedBits(I->getOperand(0), KnownZero2, KnownOne2, TD, Depth+1);
349 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
350 assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
352 // Output known-1 bits are only known if set in both the LHS & RHS.
353 KnownOne &= KnownOne2;
354 // Output known-0 are known to be clear if zero in either the LHS | RHS.
355 KnownZero |= KnownZero2;
358 case Instruction::Or: {
359 ComputeMaskedBits(I->getOperand(1), KnownZero, KnownOne, TD, Depth+1);
360 ComputeMaskedBits(I->getOperand(0), KnownZero2, KnownOne2, TD, Depth+1);
361 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
362 assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
364 // Output known-0 bits are only known if clear in both the LHS & RHS.
365 KnownZero &= KnownZero2;
366 // Output known-1 are known to be set if set in either the LHS | RHS.
367 KnownOne |= KnownOne2;
370 case Instruction::Xor: {
371 ComputeMaskedBits(I->getOperand(1), KnownZero, KnownOne, TD, Depth+1);
372 ComputeMaskedBits(I->getOperand(0), KnownZero2, KnownOne2, TD, Depth+1);
373 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
374 assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
376 // Output known-0 bits are known if clear or set in both the LHS & RHS.
377 APInt KnownZeroOut = (KnownZero & KnownZero2) | (KnownOne & KnownOne2);
378 // Output known-1 are known to be set if set in only one of the LHS, RHS.
379 KnownOne = (KnownZero & KnownOne2) | (KnownOne & KnownZero2);
380 KnownZero = KnownZeroOut;
383 case Instruction::Mul: {
384 bool NSW = cast<OverflowingBinaryOperator>(I)->hasNoSignedWrap();
385 ComputeMaskedBitsMul(I->getOperand(0), I->getOperand(1), NSW,
386 KnownZero, KnownOne, KnownZero2, KnownOne2, TD, Depth);
389 case Instruction::UDiv: {
390 // For the purposes of computing leading zeros we can conservatively
391 // treat a udiv as a logical right shift by the power of 2 known to
392 // be less than the denominator.
393 ComputeMaskedBits(I->getOperand(0), KnownZero2, KnownOne2, TD, Depth+1);
394 unsigned LeadZ = KnownZero2.countLeadingOnes();
396 KnownOne2.clearAllBits();
397 KnownZero2.clearAllBits();
398 ComputeMaskedBits(I->getOperand(1), KnownZero2, KnownOne2, TD, Depth+1);
399 unsigned RHSUnknownLeadingOnes = KnownOne2.countLeadingZeros();
400 if (RHSUnknownLeadingOnes != BitWidth)
401 LeadZ = std::min(BitWidth,
402 LeadZ + BitWidth - RHSUnknownLeadingOnes - 1);
404 KnownZero = APInt::getHighBitsSet(BitWidth, LeadZ);
407 case Instruction::Select:
408 ComputeMaskedBits(I->getOperand(2), KnownZero, KnownOne, TD, Depth+1);
409 ComputeMaskedBits(I->getOperand(1), KnownZero2, KnownOne2, TD,
411 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
412 assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
414 // Only known if known in both the LHS and RHS.
415 KnownOne &= KnownOne2;
416 KnownZero &= KnownZero2;
418 case Instruction::FPTrunc:
419 case Instruction::FPExt:
420 case Instruction::FPToUI:
421 case Instruction::FPToSI:
422 case Instruction::SIToFP:
423 case Instruction::UIToFP:
424 return; // Can't work with floating point.
425 case Instruction::PtrToInt:
426 case Instruction::IntToPtr:
427 // We can't handle these if we don't know the pointer size.
429 // FALL THROUGH and handle them the same as zext/trunc.
430 case Instruction::ZExt:
431 case Instruction::Trunc: {
432 Type *SrcTy = I->getOperand(0)->getType();
434 unsigned SrcBitWidth;
435 // Note that we handle pointer operands here because of inttoptr/ptrtoint
436 // which fall through here.
438 SrcBitWidth = TD->getTypeSizeInBits(SrcTy->getScalarType());
440 SrcBitWidth = SrcTy->getScalarSizeInBits();
441 if (!SrcBitWidth) return;
444 assert(SrcBitWidth && "SrcBitWidth can't be zero");
445 KnownZero = KnownZero.zextOrTrunc(SrcBitWidth);
446 KnownOne = KnownOne.zextOrTrunc(SrcBitWidth);
447 ComputeMaskedBits(I->getOperand(0), KnownZero, KnownOne, TD, Depth+1);
448 KnownZero = KnownZero.zextOrTrunc(BitWidth);
449 KnownOne = KnownOne.zextOrTrunc(BitWidth);
450 // Any top bits are known to be zero.
451 if (BitWidth > SrcBitWidth)
452 KnownZero |= APInt::getHighBitsSet(BitWidth, BitWidth - SrcBitWidth);
455 case Instruction::BitCast: {
456 Type *SrcTy = I->getOperand(0)->getType();
457 if ((SrcTy->isIntegerTy() || SrcTy->isPointerTy()) &&
458 // TODO: For now, not handling conversions like:
459 // (bitcast i64 %x to <2 x i32>)
460 !I->getType()->isVectorTy()) {
461 ComputeMaskedBits(I->getOperand(0), KnownZero, KnownOne, TD, Depth+1);
466 case Instruction::SExt: {
467 // Compute the bits in the result that are not present in the input.
468 unsigned SrcBitWidth = I->getOperand(0)->getType()->getScalarSizeInBits();
470 KnownZero = KnownZero.trunc(SrcBitWidth);
471 KnownOne = KnownOne.trunc(SrcBitWidth);
472 ComputeMaskedBits(I->getOperand(0), KnownZero, KnownOne, TD, Depth+1);
473 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
474 KnownZero = KnownZero.zext(BitWidth);
475 KnownOne = KnownOne.zext(BitWidth);
477 // If the sign bit of the input is known set or clear, then we know the
478 // top bits of the result.
479 if (KnownZero[SrcBitWidth-1]) // Input sign bit known zero
480 KnownZero |= APInt::getHighBitsSet(BitWidth, BitWidth - SrcBitWidth);
481 else if (KnownOne[SrcBitWidth-1]) // Input sign bit known set
482 KnownOne |= APInt::getHighBitsSet(BitWidth, BitWidth - SrcBitWidth);
485 case Instruction::Shl:
486 // (shl X, C1) & C2 == 0 iff (X & C2 >>u C1) == 0
487 if (ConstantInt *SA = dyn_cast<ConstantInt>(I->getOperand(1))) {
488 uint64_t ShiftAmt = SA->getLimitedValue(BitWidth);
489 ComputeMaskedBits(I->getOperand(0), KnownZero, KnownOne, TD, Depth+1);
490 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
491 KnownZero <<= ShiftAmt;
492 KnownOne <<= ShiftAmt;
493 KnownZero |= APInt::getLowBitsSet(BitWidth, ShiftAmt); // low bits known 0
497 case Instruction::LShr:
498 // (ushr X, C1) & C2 == 0 iff (-1 >> C1) & C2 == 0
499 if (ConstantInt *SA = dyn_cast<ConstantInt>(I->getOperand(1))) {
500 // Compute the new bits that are at the top now.
501 uint64_t ShiftAmt = SA->getLimitedValue(BitWidth);
503 // Unsigned shift right.
504 ComputeMaskedBits(I->getOperand(0), KnownZero,KnownOne, TD, Depth+1);
505 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
506 KnownZero = APIntOps::lshr(KnownZero, ShiftAmt);
507 KnownOne = APIntOps::lshr(KnownOne, ShiftAmt);
508 // high bits known zero.
509 KnownZero |= APInt::getHighBitsSet(BitWidth, ShiftAmt);
513 case Instruction::AShr:
514 // (ashr X, C1) & C2 == 0 iff (-1 >> C1) & C2 == 0
515 if (ConstantInt *SA = dyn_cast<ConstantInt>(I->getOperand(1))) {
516 // Compute the new bits that are at the top now.
517 uint64_t ShiftAmt = SA->getLimitedValue(BitWidth-1);
519 // Signed shift right.
520 ComputeMaskedBits(I->getOperand(0), KnownZero, KnownOne, TD, Depth+1);
521 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
522 KnownZero = APIntOps::lshr(KnownZero, ShiftAmt);
523 KnownOne = APIntOps::lshr(KnownOne, ShiftAmt);
525 APInt HighBits(APInt::getHighBitsSet(BitWidth, ShiftAmt));
526 if (KnownZero[BitWidth-ShiftAmt-1]) // New bits are known zero.
527 KnownZero |= HighBits;
528 else if (KnownOne[BitWidth-ShiftAmt-1]) // New bits are known one.
529 KnownOne |= HighBits;
533 case Instruction::Sub: {
534 bool NSW = cast<OverflowingBinaryOperator>(I)->hasNoSignedWrap();
535 ComputeMaskedBitsAddSub(false, I->getOperand(0), I->getOperand(1), NSW,
536 KnownZero, KnownOne, KnownZero2, KnownOne2, TD,
540 case Instruction::Add: {
541 bool NSW = cast<OverflowingBinaryOperator>(I)->hasNoSignedWrap();
542 ComputeMaskedBitsAddSub(true, I->getOperand(0), I->getOperand(1), NSW,
543 KnownZero, KnownOne, KnownZero2, KnownOne2, TD,
547 case Instruction::SRem:
548 if (ConstantInt *Rem = dyn_cast<ConstantInt>(I->getOperand(1))) {
549 APInt RA = Rem->getValue().abs();
550 if (RA.isPowerOf2()) {
551 APInt LowBits = RA - 1;
552 ComputeMaskedBits(I->getOperand(0), KnownZero2, KnownOne2, TD, Depth+1);
554 // The low bits of the first operand are unchanged by the srem.
555 KnownZero = KnownZero2 & LowBits;
556 KnownOne = KnownOne2 & LowBits;
558 // If the first operand is non-negative or has all low bits zero, then
559 // the upper bits are all zero.
560 if (KnownZero2[BitWidth-1] || ((KnownZero2 & LowBits) == LowBits))
561 KnownZero |= ~LowBits;
563 // If the first operand is negative and not all low bits are zero, then
564 // the upper bits are all one.
565 if (KnownOne2[BitWidth-1] && ((KnownOne2 & LowBits) != 0))
566 KnownOne |= ~LowBits;
568 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
572 // The sign bit is the LHS's sign bit, except when the result of the
573 // remainder is zero.
574 if (KnownZero.isNonNegative()) {
575 APInt LHSKnownZero(BitWidth, 0), LHSKnownOne(BitWidth, 0);
576 ComputeMaskedBits(I->getOperand(0), LHSKnownZero, LHSKnownOne, TD,
578 // If it's known zero, our sign bit is also zero.
579 if (LHSKnownZero.isNegative())
580 KnownZero.setBit(BitWidth - 1);
584 case Instruction::URem: {
585 if (ConstantInt *Rem = dyn_cast<ConstantInt>(I->getOperand(1))) {
586 APInt RA = Rem->getValue();
587 if (RA.isPowerOf2()) {
588 APInt LowBits = (RA - 1);
589 ComputeMaskedBits(I->getOperand(0), KnownZero, KnownOne, TD,
591 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
592 KnownZero |= ~LowBits;
598 // Since the result is less than or equal to either operand, any leading
599 // zero bits in either operand must also exist in the result.
600 ComputeMaskedBits(I->getOperand(0), KnownZero, KnownOne, TD, Depth+1);
601 ComputeMaskedBits(I->getOperand(1), KnownZero2, KnownOne2, TD, Depth+1);
603 unsigned Leaders = std::max(KnownZero.countLeadingOnes(),
604 KnownZero2.countLeadingOnes());
605 KnownOne.clearAllBits();
606 KnownZero = APInt::getHighBitsSet(BitWidth, Leaders);
610 case Instruction::Alloca: {
611 AllocaInst *AI = cast<AllocaInst>(V);
612 unsigned Align = AI->getAlignment();
613 if (Align == 0 && TD)
614 Align = TD->getABITypeAlignment(AI->getType()->getElementType());
617 KnownZero = APInt::getLowBitsSet(BitWidth, countTrailingZeros(Align));
620 case Instruction::GetElementPtr: {
621 // Analyze all of the subscripts of this getelementptr instruction
622 // to determine if we can prove known low zero bits.
623 APInt LocalKnownZero(BitWidth, 0), LocalKnownOne(BitWidth, 0);
624 ComputeMaskedBits(I->getOperand(0), LocalKnownZero, LocalKnownOne, TD,
626 unsigned TrailZ = LocalKnownZero.countTrailingOnes();
628 gep_type_iterator GTI = gep_type_begin(I);
629 for (unsigned i = 1, e = I->getNumOperands(); i != e; ++i, ++GTI) {
630 Value *Index = I->getOperand(i);
631 if (StructType *STy = dyn_cast<StructType>(*GTI)) {
632 // Handle struct member offset arithmetic.
636 // Handle case when index is vector zeroinitializer
637 Constant *CIndex = cast<Constant>(Index);
638 if (CIndex->isZeroValue())
641 if (CIndex->getType()->isVectorTy())
642 Index = CIndex->getSplatValue();
644 unsigned Idx = cast<ConstantInt>(Index)->getZExtValue();
645 const StructLayout *SL = TD->getStructLayout(STy);
646 uint64_t Offset = SL->getElementOffset(Idx);
647 TrailZ = std::min<unsigned>(TrailZ,
648 countTrailingZeros(Offset));
650 // Handle array index arithmetic.
651 Type *IndexedTy = GTI.getIndexedType();
652 if (!IndexedTy->isSized()) return;
653 unsigned GEPOpiBits = Index->getType()->getScalarSizeInBits();
654 uint64_t TypeSize = TD ? TD->getTypeAllocSize(IndexedTy) : 1;
655 LocalKnownZero = LocalKnownOne = APInt(GEPOpiBits, 0);
656 ComputeMaskedBits(Index, LocalKnownZero, LocalKnownOne, TD, Depth+1);
657 TrailZ = std::min(TrailZ,
658 unsigned(countTrailingZeros(TypeSize) +
659 LocalKnownZero.countTrailingOnes()));
663 KnownZero = APInt::getLowBitsSet(BitWidth, TrailZ);
666 case Instruction::PHI: {
667 PHINode *P = cast<PHINode>(I);
668 // Handle the case of a simple two-predecessor recurrence PHI.
669 // There's a lot more that could theoretically be done here, but
670 // this is sufficient to catch some interesting cases.
671 if (P->getNumIncomingValues() == 2) {
672 for (unsigned i = 0; i != 2; ++i) {
673 Value *L = P->getIncomingValue(i);
674 Value *R = P->getIncomingValue(!i);
675 Operator *LU = dyn_cast<Operator>(L);
678 unsigned Opcode = LU->getOpcode();
679 // Check for operations that have the property that if
680 // both their operands have low zero bits, the result
681 // will have low zero bits.
682 if (Opcode == Instruction::Add ||
683 Opcode == Instruction::Sub ||
684 Opcode == Instruction::And ||
685 Opcode == Instruction::Or ||
686 Opcode == Instruction::Mul) {
687 Value *LL = LU->getOperand(0);
688 Value *LR = LU->getOperand(1);
689 // Find a recurrence.
696 // Ok, we have a PHI of the form L op= R. Check for low
698 ComputeMaskedBits(R, KnownZero2, KnownOne2, TD, Depth+1);
700 // We need to take the minimum number of known bits
701 APInt KnownZero3(KnownZero), KnownOne3(KnownOne);
702 ComputeMaskedBits(L, KnownZero3, KnownOne3, TD, Depth+1);
704 KnownZero = APInt::getLowBitsSet(BitWidth,
705 std::min(KnownZero2.countTrailingOnes(),
706 KnownZero3.countTrailingOnes()));
712 // Unreachable blocks may have zero-operand PHI nodes.
713 if (P->getNumIncomingValues() == 0)
716 // Otherwise take the unions of the known bit sets of the operands,
717 // taking conservative care to avoid excessive recursion.
718 if (Depth < MaxDepth - 1 && !KnownZero && !KnownOne) {
719 // Skip if every incoming value references to ourself.
720 if (dyn_cast_or_null<UndefValue>(P->hasConstantValue()))
723 KnownZero = APInt::getAllOnesValue(BitWidth);
724 KnownOne = APInt::getAllOnesValue(BitWidth);
725 for (unsigned i = 0, e = P->getNumIncomingValues(); i != e; ++i) {
726 // Skip direct self references.
727 if (P->getIncomingValue(i) == P) continue;
729 KnownZero2 = APInt(BitWidth, 0);
730 KnownOne2 = APInt(BitWidth, 0);
731 // Recurse, but cap the recursion to one level, because we don't
732 // want to waste time spinning around in loops.
733 ComputeMaskedBits(P->getIncomingValue(i), KnownZero2, KnownOne2, TD,
735 KnownZero &= KnownZero2;
736 KnownOne &= KnownOne2;
737 // If all bits have been ruled out, there's no need to check
739 if (!KnownZero && !KnownOne)
745 case Instruction::Call:
746 if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
747 switch (II->getIntrinsicID()) {
749 case Intrinsic::ctlz:
750 case Intrinsic::cttz: {
751 unsigned LowBits = Log2_32(BitWidth)+1;
752 // If this call is undefined for 0, the result will be less than 2^n.
753 if (II->getArgOperand(1) == ConstantInt::getTrue(II->getContext()))
755 KnownZero = APInt::getHighBitsSet(BitWidth, BitWidth - LowBits);
758 case Intrinsic::ctpop: {
759 unsigned LowBits = Log2_32(BitWidth)+1;
760 KnownZero = APInt::getHighBitsSet(BitWidth, BitWidth - LowBits);
763 case Intrinsic::x86_sse42_crc32_64_64:
764 KnownZero = APInt::getHighBitsSet(64, 32);
769 case Instruction::ExtractValue:
770 if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I->getOperand(0))) {
771 ExtractValueInst *EVI = cast<ExtractValueInst>(I);
772 if (EVI->getNumIndices() != 1) break;
773 if (EVI->getIndices()[0] == 0) {
774 switch (II->getIntrinsicID()) {
776 case Intrinsic::uadd_with_overflow:
777 case Intrinsic::sadd_with_overflow:
778 ComputeMaskedBitsAddSub(true, II->getArgOperand(0),
779 II->getArgOperand(1), false, KnownZero,
780 KnownOne, KnownZero2, KnownOne2, TD, Depth);
782 case Intrinsic::usub_with_overflow:
783 case Intrinsic::ssub_with_overflow:
784 ComputeMaskedBitsAddSub(false, II->getArgOperand(0),
785 II->getArgOperand(1), false, KnownZero,
786 KnownOne, KnownZero2, KnownOne2, TD, Depth);
788 case Intrinsic::umul_with_overflow:
789 case Intrinsic::smul_with_overflow:
790 ComputeMaskedBitsMul(II->getArgOperand(0), II->getArgOperand(1),
791 false, KnownZero, KnownOne,
792 KnownZero2, KnownOne2, TD, Depth);
800 /// ComputeSignBit - Determine whether the sign bit is known to be zero or
801 /// one. Convenience wrapper around ComputeMaskedBits.
802 void llvm::ComputeSignBit(Value *V, bool &KnownZero, bool &KnownOne,
803 const DataLayout *TD, unsigned Depth) {
804 unsigned BitWidth = getBitWidth(V->getType(), TD);
810 APInt ZeroBits(BitWidth, 0);
811 APInt OneBits(BitWidth, 0);
812 ComputeMaskedBits(V, ZeroBits, OneBits, TD, Depth);
813 KnownOne = OneBits[BitWidth - 1];
814 KnownZero = ZeroBits[BitWidth - 1];
817 /// isKnownToBeAPowerOfTwo - Return true if the given value is known to have exactly one
818 /// bit set when defined. For vectors return true if every element is known to
819 /// be a power of two when defined. Supports values with integer or pointer
820 /// types and vectors of integers.
821 bool llvm::isKnownToBeAPowerOfTwo(Value *V, bool OrZero, unsigned Depth) {
822 if (Constant *C = dyn_cast<Constant>(V)) {
823 if (C->isNullValue())
825 if (ConstantInt *CI = dyn_cast<ConstantInt>(C))
826 return CI->getValue().isPowerOf2();
827 // TODO: Handle vector constants.
830 // 1 << X is clearly a power of two if the one is not shifted off the end. If
831 // it is shifted off the end then the result is undefined.
832 if (match(V, m_Shl(m_One(), m_Value())))
835 // (signbit) >>l X is clearly a power of two if the one is not shifted off the
836 // bottom. If it is shifted off the bottom then the result is undefined.
837 if (match(V, m_LShr(m_SignBit(), m_Value())))
840 // The remaining tests are all recursive, so bail out if we hit the limit.
841 if (Depth++ == MaxDepth)
844 Value *X = 0, *Y = 0;
845 // A shift of a power of two is a power of two or zero.
846 if (OrZero && (match(V, m_Shl(m_Value(X), m_Value())) ||
847 match(V, m_Shr(m_Value(X), m_Value()))))
848 return isKnownToBeAPowerOfTwo(X, /*OrZero*/true, Depth);
850 if (ZExtInst *ZI = dyn_cast<ZExtInst>(V))
851 return isKnownToBeAPowerOfTwo(ZI->getOperand(0), OrZero, Depth);
853 if (SelectInst *SI = dyn_cast<SelectInst>(V))
854 return isKnownToBeAPowerOfTwo(SI->getTrueValue(), OrZero, Depth) &&
855 isKnownToBeAPowerOfTwo(SI->getFalseValue(), OrZero, Depth);
857 if (OrZero && match(V, m_And(m_Value(X), m_Value(Y)))) {
858 // A power of two and'd with anything is a power of two or zero.
859 if (isKnownToBeAPowerOfTwo(X, /*OrZero*/true, Depth) ||
860 isKnownToBeAPowerOfTwo(Y, /*OrZero*/true, Depth))
862 // X & (-X) is always a power of two or zero.
863 if (match(X, m_Neg(m_Specific(Y))) || match(Y, m_Neg(m_Specific(X))))
868 // Adding a power-of-two or zero to the same power-of-two or zero yields
869 // either the original power-of-two, a larger power-of-two or zero.
870 if (match(V, m_Add(m_Value(X), m_Value(Y)))) {
871 OverflowingBinaryOperator *VOBO = cast<OverflowingBinaryOperator>(V);
872 if (OrZero || VOBO->hasNoUnsignedWrap() || VOBO->hasNoSignedWrap()) {
873 if (match(X, m_And(m_Specific(Y), m_Value())) ||
874 match(X, m_And(m_Value(), m_Specific(Y))))
875 if (isKnownToBeAPowerOfTwo(Y, OrZero, Depth))
877 if (match(Y, m_And(m_Specific(X), m_Value())) ||
878 match(Y, m_And(m_Value(), m_Specific(X))))
879 if (isKnownToBeAPowerOfTwo(X, OrZero, Depth))
882 unsigned BitWidth = V->getType()->getScalarSizeInBits();
883 APInt LHSZeroBits(BitWidth, 0), LHSOneBits(BitWidth, 0);
884 ComputeMaskedBits(X, LHSZeroBits, LHSOneBits, 0, Depth);
886 APInt RHSZeroBits(BitWidth, 0), RHSOneBits(BitWidth, 0);
887 ComputeMaskedBits(Y, RHSZeroBits, RHSOneBits, 0, Depth);
888 // If i8 V is a power of two or zero:
889 // ZeroBits: 1 1 1 0 1 1 1 1
890 // ~ZeroBits: 0 0 0 1 0 0 0 0
891 if ((~(LHSZeroBits & RHSZeroBits)).isPowerOf2())
892 // If OrZero isn't set, we cannot give back a zero result.
893 // Make sure either the LHS or RHS has a bit set.
894 if (OrZero || RHSOneBits.getBoolValue() || LHSOneBits.getBoolValue())
899 // An exact divide or right shift can only shift off zero bits, so the result
900 // is a power of two only if the first operand is a power of two and not
901 // copying a sign bit (sdiv int_min, 2).
902 if (match(V, m_Exact(m_LShr(m_Value(), m_Value()))) ||
903 match(V, m_Exact(m_UDiv(m_Value(), m_Value())))) {
904 return isKnownToBeAPowerOfTwo(cast<Operator>(V)->getOperand(0), OrZero, Depth);
910 /// \brief Test whether a GEP's result is known to be non-null.
912 /// Uses properties inherent in a GEP to try to determine whether it is known
915 /// Currently this routine does not support vector GEPs.
916 static bool isGEPKnownNonNull(GEPOperator *GEP, const DataLayout *DL,
918 if (!GEP->isInBounds() || GEP->getPointerAddressSpace() != 0)
921 // FIXME: Support vector-GEPs.
922 assert(GEP->getType()->isPointerTy() && "We only support plain pointer GEP");
924 // If the base pointer is non-null, we cannot walk to a null address with an
925 // inbounds GEP in address space zero.
926 if (isKnownNonZero(GEP->getPointerOperand(), DL, Depth))
929 // Past this, if we don't have DataLayout, we can't do much.
933 // Walk the GEP operands and see if any operand introduces a non-zero offset.
934 // If so, then the GEP cannot produce a null pointer, as doing so would
935 // inherently violate the inbounds contract within address space zero.
936 for (gep_type_iterator GTI = gep_type_begin(GEP), GTE = gep_type_end(GEP);
938 // Struct types are easy -- they must always be indexed by a constant.
939 if (StructType *STy = dyn_cast<StructType>(*GTI)) {
940 ConstantInt *OpC = cast<ConstantInt>(GTI.getOperand());
941 unsigned ElementIdx = OpC->getZExtValue();
942 const StructLayout *SL = DL->getStructLayout(STy);
943 uint64_t ElementOffset = SL->getElementOffset(ElementIdx);
944 if (ElementOffset > 0)
949 // If we have a zero-sized type, the index doesn't matter. Keep looping.
950 if (DL->getTypeAllocSize(GTI.getIndexedType()) == 0)
953 // Fast path the constant operand case both for efficiency and so we don't
954 // increment Depth when just zipping down an all-constant GEP.
955 if (ConstantInt *OpC = dyn_cast<ConstantInt>(GTI.getOperand())) {
961 // We post-increment Depth here because while isKnownNonZero increments it
962 // as well, when we pop back up that increment won't persist. We don't want
963 // to recurse 10k times just because we have 10k GEP operands. We don't
964 // bail completely out because we want to handle constant GEPs regardless
966 if (Depth++ >= MaxDepth)
969 if (isKnownNonZero(GTI.getOperand(), DL, Depth))
976 /// isKnownNonZero - Return true if the given value is known to be non-zero
977 /// when defined. For vectors return true if every element is known to be
978 /// non-zero when defined. Supports values with integer or pointer type and
979 /// vectors of integers.
980 bool llvm::isKnownNonZero(Value *V, const DataLayout *TD, unsigned Depth) {
981 if (Constant *C = dyn_cast<Constant>(V)) {
982 if (C->isNullValue())
984 if (isa<ConstantInt>(C))
985 // Must be non-zero due to null test above.
987 // TODO: Handle vectors
991 // The remaining tests are all recursive, so bail out if we hit the limit.
992 if (Depth++ >= MaxDepth)
995 // Check for pointer simplifications.
996 if (V->getType()->isPointerTy()) {
997 if (isKnownNonNull(V))
999 if (GEPOperator *GEP = dyn_cast<GEPOperator>(V))
1000 if (isGEPKnownNonNull(GEP, TD, Depth))
1004 unsigned BitWidth = getBitWidth(V->getType()->getScalarType(), TD);
1006 // X | Y != 0 if X != 0 or Y != 0.
1007 Value *X = 0, *Y = 0;
1008 if (match(V, m_Or(m_Value(X), m_Value(Y))))
1009 return isKnownNonZero(X, TD, Depth) || isKnownNonZero(Y, TD, Depth);
1011 // ext X != 0 if X != 0.
1012 if (isa<SExtInst>(V) || isa<ZExtInst>(V))
1013 return isKnownNonZero(cast<Instruction>(V)->getOperand(0), TD, Depth);
1015 // shl X, Y != 0 if X is odd. Note that the value of the shift is undefined
1016 // if the lowest bit is shifted off the end.
1017 if (BitWidth && match(V, m_Shl(m_Value(X), m_Value(Y)))) {
1018 // shl nuw can't remove any non-zero bits.
1019 OverflowingBinaryOperator *BO = cast<OverflowingBinaryOperator>(V);
1020 if (BO->hasNoUnsignedWrap())
1021 return isKnownNonZero(X, TD, Depth);
1023 APInt KnownZero(BitWidth, 0);
1024 APInt KnownOne(BitWidth, 0);
1025 ComputeMaskedBits(X, KnownZero, KnownOne, TD, Depth);
1029 // shr X, Y != 0 if X is negative. Note that the value of the shift is not
1030 // defined if the sign bit is shifted off the end.
1031 else if (match(V, m_Shr(m_Value(X), m_Value(Y)))) {
1032 // shr exact can only shift out zero bits.
1033 PossiblyExactOperator *BO = cast<PossiblyExactOperator>(V);
1035 return isKnownNonZero(X, TD, Depth);
1037 bool XKnownNonNegative, XKnownNegative;
1038 ComputeSignBit(X, XKnownNonNegative, XKnownNegative, TD, Depth);
1042 // div exact can only produce a zero if the dividend is zero.
1043 else if (match(V, m_Exact(m_IDiv(m_Value(X), m_Value())))) {
1044 return isKnownNonZero(X, TD, Depth);
1047 else if (match(V, m_Add(m_Value(X), m_Value(Y)))) {
1048 bool XKnownNonNegative, XKnownNegative;
1049 bool YKnownNonNegative, YKnownNegative;
1050 ComputeSignBit(X, XKnownNonNegative, XKnownNegative, TD, Depth);
1051 ComputeSignBit(Y, YKnownNonNegative, YKnownNegative, TD, Depth);
1053 // If X and Y are both non-negative (as signed values) then their sum is not
1054 // zero unless both X and Y are zero.
1055 if (XKnownNonNegative && YKnownNonNegative)
1056 if (isKnownNonZero(X, TD, Depth) || isKnownNonZero(Y, TD, Depth))
1059 // If X and Y are both negative (as signed values) then their sum is not
1060 // zero unless both X and Y equal INT_MIN.
1061 if (BitWidth && XKnownNegative && YKnownNegative) {
1062 APInt KnownZero(BitWidth, 0);
1063 APInt KnownOne(BitWidth, 0);
1064 APInt Mask = APInt::getSignedMaxValue(BitWidth);
1065 // The sign bit of X is set. If some other bit is set then X is not equal
1067 ComputeMaskedBits(X, KnownZero, KnownOne, TD, Depth);
1068 if ((KnownOne & Mask) != 0)
1070 // The sign bit of Y is set. If some other bit is set then Y is not equal
1072 ComputeMaskedBits(Y, KnownZero, KnownOne, TD, Depth);
1073 if ((KnownOne & Mask) != 0)
1077 // The sum of a non-negative number and a power of two is not zero.
1078 if (XKnownNonNegative && isKnownToBeAPowerOfTwo(Y, /*OrZero*/false, Depth))
1080 if (YKnownNonNegative && isKnownToBeAPowerOfTwo(X, /*OrZero*/false, Depth))
1084 else if (match(V, m_Mul(m_Value(X), m_Value(Y)))) {
1085 OverflowingBinaryOperator *BO = cast<OverflowingBinaryOperator>(V);
1086 // If X and Y are non-zero then so is X * Y as long as the multiplication
1087 // does not overflow.
1088 if ((BO->hasNoSignedWrap() || BO->hasNoUnsignedWrap()) &&
1089 isKnownNonZero(X, TD, Depth) && isKnownNonZero(Y, TD, Depth))
1092 // (C ? X : Y) != 0 if X != 0 and Y != 0.
1093 else if (SelectInst *SI = dyn_cast<SelectInst>(V)) {
1094 if (isKnownNonZero(SI->getTrueValue(), TD, Depth) &&
1095 isKnownNonZero(SI->getFalseValue(), TD, Depth))
1099 if (!BitWidth) return false;
1100 APInt KnownZero(BitWidth, 0);
1101 APInt KnownOne(BitWidth, 0);
1102 ComputeMaskedBits(V, KnownZero, KnownOne, TD, Depth);
1103 return KnownOne != 0;
1106 /// MaskedValueIsZero - Return true if 'V & Mask' is known to be zero. We use
1107 /// this predicate to simplify operations downstream. Mask is known to be zero
1108 /// for bits that V cannot have.
1110 /// This function is defined on values with integer type, values with pointer
1111 /// type (but only if TD is non-null), and vectors of integers. In the case
1112 /// where V is a vector, the mask, known zero, and known one values are the
1113 /// same width as the vector element, and the bit is set only if it is true
1114 /// for all of the elements in the vector.
1115 bool llvm::MaskedValueIsZero(Value *V, const APInt &Mask,
1116 const DataLayout *TD, unsigned Depth) {
1117 APInt KnownZero(Mask.getBitWidth(), 0), KnownOne(Mask.getBitWidth(), 0);
1118 ComputeMaskedBits(V, KnownZero, KnownOne, TD, Depth);
1119 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
1120 return (KnownZero & Mask) == Mask;
1125 /// ComputeNumSignBits - Return the number of times the sign bit of the
1126 /// register is replicated into the other bits. We know that at least 1 bit
1127 /// is always equal to the sign bit (itself), but other cases can give us
1128 /// information. For example, immediately after an "ashr X, 2", we know that
1129 /// the top 3 bits are all equal to each other, so we return 3.
1131 /// 'Op' must have a scalar integer type.
1133 unsigned llvm::ComputeNumSignBits(Value *V, const DataLayout *TD,
1135 assert((TD || V->getType()->isIntOrIntVectorTy()) &&
1136 "ComputeNumSignBits requires a DataLayout object to operate "
1137 "on non-integer values!");
1138 Type *Ty = V->getType();
1139 unsigned TyBits = TD ? TD->getTypeSizeInBits(V->getType()->getScalarType()) :
1140 Ty->getScalarSizeInBits();
1142 unsigned FirstAnswer = 1;
1144 // Note that ConstantInt is handled by the general ComputeMaskedBits case
1148 return 1; // Limit search depth.
1150 Operator *U = dyn_cast<Operator>(V);
1151 switch (Operator::getOpcode(V)) {
1153 case Instruction::SExt:
1154 Tmp = TyBits - U->getOperand(0)->getType()->getScalarSizeInBits();
1155 return ComputeNumSignBits(U->getOperand(0), TD, Depth+1) + Tmp;
1157 case Instruction::AShr: {
1158 Tmp = ComputeNumSignBits(U->getOperand(0), TD, Depth+1);
1159 // ashr X, C -> adds C sign bits. Vectors too.
1161 if (match(U->getOperand(1), m_APInt(ShAmt))) {
1162 Tmp += ShAmt->getZExtValue();
1163 if (Tmp > TyBits) Tmp = TyBits;
1167 case Instruction::Shl: {
1169 if (match(U->getOperand(1), m_APInt(ShAmt))) {
1170 // shl destroys sign bits.
1171 Tmp = ComputeNumSignBits(U->getOperand(0), TD, Depth+1);
1172 Tmp2 = ShAmt->getZExtValue();
1173 if (Tmp2 >= TyBits || // Bad shift.
1174 Tmp2 >= Tmp) break; // Shifted all sign bits out.
1179 case Instruction::And:
1180 case Instruction::Or:
1181 case Instruction::Xor: // NOT is handled here.
1182 // Logical binary ops preserve the number of sign bits at the worst.
1183 Tmp = ComputeNumSignBits(U->getOperand(0), TD, Depth+1);
1185 Tmp2 = ComputeNumSignBits(U->getOperand(1), TD, Depth+1);
1186 FirstAnswer = std::min(Tmp, Tmp2);
1187 // We computed what we know about the sign bits as our first
1188 // answer. Now proceed to the generic code that uses
1189 // ComputeMaskedBits, and pick whichever answer is better.
1193 case Instruction::Select:
1194 Tmp = ComputeNumSignBits(U->getOperand(1), TD, Depth+1);
1195 if (Tmp == 1) return 1; // Early out.
1196 Tmp2 = ComputeNumSignBits(U->getOperand(2), TD, Depth+1);
1197 return std::min(Tmp, Tmp2);
1199 case Instruction::Add:
1200 // Add can have at most one carry bit. Thus we know that the output
1201 // is, at worst, one more bit than the inputs.
1202 Tmp = ComputeNumSignBits(U->getOperand(0), TD, Depth+1);
1203 if (Tmp == 1) return 1; // Early out.
1205 // Special case decrementing a value (ADD X, -1):
1206 if (ConstantInt *CRHS = dyn_cast<ConstantInt>(U->getOperand(1)))
1207 if (CRHS->isAllOnesValue()) {
1208 APInt KnownZero(TyBits, 0), KnownOne(TyBits, 0);
1209 ComputeMaskedBits(U->getOperand(0), KnownZero, KnownOne, TD, Depth+1);
1211 // If the input is known to be 0 or 1, the output is 0/-1, which is all
1213 if ((KnownZero | APInt(TyBits, 1)).isAllOnesValue())
1216 // If we are subtracting one from a positive number, there is no carry
1217 // out of the result.
1218 if (KnownZero.isNegative())
1222 Tmp2 = ComputeNumSignBits(U->getOperand(1), TD, Depth+1);
1223 if (Tmp2 == 1) return 1;
1224 return std::min(Tmp, Tmp2)-1;
1226 case Instruction::Sub:
1227 Tmp2 = ComputeNumSignBits(U->getOperand(1), TD, Depth+1);
1228 if (Tmp2 == 1) return 1;
1231 if (ConstantInt *CLHS = dyn_cast<ConstantInt>(U->getOperand(0)))
1232 if (CLHS->isNullValue()) {
1233 APInt KnownZero(TyBits, 0), KnownOne(TyBits, 0);
1234 ComputeMaskedBits(U->getOperand(1), KnownZero, KnownOne, TD, Depth+1);
1235 // If the input is known to be 0 or 1, the output is 0/-1, which is all
1237 if ((KnownZero | APInt(TyBits, 1)).isAllOnesValue())
1240 // If the input is known to be positive (the sign bit is known clear),
1241 // the output of the NEG has the same number of sign bits as the input.
1242 if (KnownZero.isNegative())
1245 // Otherwise, we treat this like a SUB.
1248 // Sub can have at most one carry bit. Thus we know that the output
1249 // is, at worst, one more bit than the inputs.
1250 Tmp = ComputeNumSignBits(U->getOperand(0), TD, Depth+1);
1251 if (Tmp == 1) return 1; // Early out.
1252 return std::min(Tmp, Tmp2)-1;
1254 case Instruction::PHI: {
1255 PHINode *PN = cast<PHINode>(U);
1256 // Don't analyze large in-degree PHIs.
1257 if (PN->getNumIncomingValues() > 4) break;
1259 // Take the minimum of all incoming values. This can't infinitely loop
1260 // because of our depth threshold.
1261 Tmp = ComputeNumSignBits(PN->getIncomingValue(0), TD, Depth+1);
1262 for (unsigned i = 1, e = PN->getNumIncomingValues(); i != e; ++i) {
1263 if (Tmp == 1) return Tmp;
1265 ComputeNumSignBits(PN->getIncomingValue(i), TD, Depth+1));
1270 case Instruction::Trunc:
1271 // FIXME: it's tricky to do anything useful for this, but it is an important
1272 // case for targets like X86.
1276 // Finally, if we can prove that the top bits of the result are 0's or 1's,
1277 // use this information.
1278 APInt KnownZero(TyBits, 0), KnownOne(TyBits, 0);
1280 ComputeMaskedBits(V, KnownZero, KnownOne, TD, Depth);
1282 if (KnownZero.isNegative()) { // sign bit is 0
1284 } else if (KnownOne.isNegative()) { // sign bit is 1;
1291 // Okay, we know that the sign bit in Mask is set. Use CLZ to determine
1292 // the number of identical bits in the top of the input value.
1294 Mask <<= Mask.getBitWidth()-TyBits;
1295 // Return # leading zeros. We use 'min' here in case Val was zero before
1296 // shifting. We don't want to return '64' as for an i32 "0".
1297 return std::max(FirstAnswer, std::min(TyBits, Mask.countLeadingZeros()));
1300 /// ComputeMultiple - This function computes the integer multiple of Base that
1301 /// equals V. If successful, it returns true and returns the multiple in
1302 /// Multiple. If unsuccessful, it returns false. It looks
1303 /// through SExt instructions only if LookThroughSExt is true.
1304 bool llvm::ComputeMultiple(Value *V, unsigned Base, Value *&Multiple,
1305 bool LookThroughSExt, unsigned Depth) {
1306 const unsigned MaxDepth = 6;
1308 assert(V && "No Value?");
1309 assert(Depth <= MaxDepth && "Limit Search Depth");
1310 assert(V->getType()->isIntegerTy() && "Not integer or pointer type!");
1312 Type *T = V->getType();
1314 ConstantInt *CI = dyn_cast<ConstantInt>(V);
1324 ConstantExpr *CO = dyn_cast<ConstantExpr>(V);
1325 Constant *BaseVal = ConstantInt::get(T, Base);
1326 if (CO && CO == BaseVal) {
1328 Multiple = ConstantInt::get(T, 1);
1332 if (CI && CI->getZExtValue() % Base == 0) {
1333 Multiple = ConstantInt::get(T, CI->getZExtValue() / Base);
1337 if (Depth == MaxDepth) return false; // Limit search depth.
1339 Operator *I = dyn_cast<Operator>(V);
1340 if (!I) return false;
1342 switch (I->getOpcode()) {
1344 case Instruction::SExt:
1345 if (!LookThroughSExt) return false;
1346 // otherwise fall through to ZExt
1347 case Instruction::ZExt:
1348 return ComputeMultiple(I->getOperand(0), Base, Multiple,
1349 LookThroughSExt, Depth+1);
1350 case Instruction::Shl:
1351 case Instruction::Mul: {
1352 Value *Op0 = I->getOperand(0);
1353 Value *Op1 = I->getOperand(1);
1355 if (I->getOpcode() == Instruction::Shl) {
1356 ConstantInt *Op1CI = dyn_cast<ConstantInt>(Op1);
1357 if (!Op1CI) return false;
1358 // Turn Op0 << Op1 into Op0 * 2^Op1
1359 APInt Op1Int = Op1CI->getValue();
1360 uint64_t BitToSet = Op1Int.getLimitedValue(Op1Int.getBitWidth() - 1);
1361 APInt API(Op1Int.getBitWidth(), 0);
1362 API.setBit(BitToSet);
1363 Op1 = ConstantInt::get(V->getContext(), API);
1367 if (ComputeMultiple(Op0, Base, Mul0, LookThroughSExt, Depth+1)) {
1368 if (Constant *Op1C = dyn_cast<Constant>(Op1))
1369 if (Constant *MulC = dyn_cast<Constant>(Mul0)) {
1370 if (Op1C->getType()->getPrimitiveSizeInBits() <
1371 MulC->getType()->getPrimitiveSizeInBits())
1372 Op1C = ConstantExpr::getZExt(Op1C, MulC->getType());
1373 if (Op1C->getType()->getPrimitiveSizeInBits() >
1374 MulC->getType()->getPrimitiveSizeInBits())
1375 MulC = ConstantExpr::getZExt(MulC, Op1C->getType());
1377 // V == Base * (Mul0 * Op1), so return (Mul0 * Op1)
1378 Multiple = ConstantExpr::getMul(MulC, Op1C);
1382 if (ConstantInt *Mul0CI = dyn_cast<ConstantInt>(Mul0))
1383 if (Mul0CI->getValue() == 1) {
1384 // V == Base * Op1, so return Op1
1391 if (ComputeMultiple(Op1, Base, Mul1, LookThroughSExt, Depth+1)) {
1392 if (Constant *Op0C = dyn_cast<Constant>(Op0))
1393 if (Constant *MulC = dyn_cast<Constant>(Mul1)) {
1394 if (Op0C->getType()->getPrimitiveSizeInBits() <
1395 MulC->getType()->getPrimitiveSizeInBits())
1396 Op0C = ConstantExpr::getZExt(Op0C, MulC->getType());
1397 if (Op0C->getType()->getPrimitiveSizeInBits() >
1398 MulC->getType()->getPrimitiveSizeInBits())
1399 MulC = ConstantExpr::getZExt(MulC, Op0C->getType());
1401 // V == Base * (Mul1 * Op0), so return (Mul1 * Op0)
1402 Multiple = ConstantExpr::getMul(MulC, Op0C);
1406 if (ConstantInt *Mul1CI = dyn_cast<ConstantInt>(Mul1))
1407 if (Mul1CI->getValue() == 1) {
1408 // V == Base * Op0, so return Op0
1416 // We could not determine if V is a multiple of Base.
1420 /// CannotBeNegativeZero - Return true if we can prove that the specified FP
1421 /// value is never equal to -0.0.
1423 /// NOTE: this function will need to be revisited when we support non-default
1426 bool llvm::CannotBeNegativeZero(const Value *V, unsigned Depth) {
1427 if (const ConstantFP *CFP = dyn_cast<ConstantFP>(V))
1428 return !CFP->getValueAPF().isNegZero();
1431 return 1; // Limit search depth.
1433 const Operator *I = dyn_cast<Operator>(V);
1434 if (I == 0) return false;
1436 // Check if the nsz fast-math flag is set
1437 if (const FPMathOperator *FPO = dyn_cast<FPMathOperator>(I))
1438 if (FPO->hasNoSignedZeros())
1441 // (add x, 0.0) is guaranteed to return +0.0, not -0.0.
1442 if (I->getOpcode() == Instruction::FAdd)
1443 if (ConstantFP *CFP = dyn_cast<ConstantFP>(I->getOperand(1)))
1444 if (CFP->isNullValue())
1447 // sitofp and uitofp turn into +0.0 for zero.
1448 if (isa<SIToFPInst>(I) || isa<UIToFPInst>(I))
1451 if (const IntrinsicInst *II = dyn_cast<IntrinsicInst>(I))
1452 // sqrt(-0.0) = -0.0, no other negative results are possible.
1453 if (II->getIntrinsicID() == Intrinsic::sqrt)
1454 return CannotBeNegativeZero(II->getArgOperand(0), Depth+1);
1456 if (const CallInst *CI = dyn_cast<CallInst>(I))
1457 if (const Function *F = CI->getCalledFunction()) {
1458 if (F->isDeclaration()) {
1460 if (F->getName() == "abs") return true;
1461 // fabs[lf](x) != -0.0
1462 if (F->getName() == "fabs") return true;
1463 if (F->getName() == "fabsf") return true;
1464 if (F->getName() == "fabsl") return true;
1465 if (F->getName() == "sqrt" || F->getName() == "sqrtf" ||
1466 F->getName() == "sqrtl")
1467 return CannotBeNegativeZero(CI->getArgOperand(0), Depth+1);
1474 /// isBytewiseValue - If the specified value can be set by repeating the same
1475 /// byte in memory, return the i8 value that it is represented with. This is
1476 /// true for all i8 values obviously, but is also true for i32 0, i32 -1,
1477 /// i16 0xF0F0, double 0.0 etc. If the value can't be handled with a repeated
1478 /// byte store (e.g. i16 0x1234), return null.
1479 Value *llvm::isBytewiseValue(Value *V) {
1480 // All byte-wide stores are splatable, even of arbitrary variables.
1481 if (V->getType()->isIntegerTy(8)) return V;
1483 // Handle 'null' ConstantArrayZero etc.
1484 if (Constant *C = dyn_cast<Constant>(V))
1485 if (C->isNullValue())
1486 return Constant::getNullValue(Type::getInt8Ty(V->getContext()));
1488 // Constant float and double values can be handled as integer values if the
1489 // corresponding integer value is "byteable". An important case is 0.0.
1490 if (ConstantFP *CFP = dyn_cast<ConstantFP>(V)) {
1491 if (CFP->getType()->isFloatTy())
1492 V = ConstantExpr::getBitCast(CFP, Type::getInt32Ty(V->getContext()));
1493 if (CFP->getType()->isDoubleTy())
1494 V = ConstantExpr::getBitCast(CFP, Type::getInt64Ty(V->getContext()));
1495 // Don't handle long double formats, which have strange constraints.
1498 // We can handle constant integers that are power of two in size and a
1499 // multiple of 8 bits.
1500 if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
1501 unsigned Width = CI->getBitWidth();
1502 if (isPowerOf2_32(Width) && Width > 8) {
1503 // We can handle this value if the recursive binary decomposition is the
1504 // same at all levels.
1505 APInt Val = CI->getValue();
1507 while (Val.getBitWidth() != 8) {
1508 unsigned NextWidth = Val.getBitWidth()/2;
1509 Val2 = Val.lshr(NextWidth);
1510 Val2 = Val2.trunc(Val.getBitWidth()/2);
1511 Val = Val.trunc(Val.getBitWidth()/2);
1513 // If the top/bottom halves aren't the same, reject it.
1517 return ConstantInt::get(V->getContext(), Val);
1521 // A ConstantDataArray/Vector is splatable if all its members are equal and
1523 if (ConstantDataSequential *CA = dyn_cast<ConstantDataSequential>(V)) {
1524 Value *Elt = CA->getElementAsConstant(0);
1525 Value *Val = isBytewiseValue(Elt);
1529 for (unsigned I = 1, E = CA->getNumElements(); I != E; ++I)
1530 if (CA->getElementAsConstant(I) != Elt)
1536 // Conceptually, we could handle things like:
1537 // %a = zext i8 %X to i16
1538 // %b = shl i16 %a, 8
1539 // %c = or i16 %a, %b
1540 // but until there is an example that actually needs this, it doesn't seem
1541 // worth worrying about.
1546 // This is the recursive version of BuildSubAggregate. It takes a few different
1547 // arguments. Idxs is the index within the nested struct From that we are
1548 // looking at now (which is of type IndexedType). IdxSkip is the number of
1549 // indices from Idxs that should be left out when inserting into the resulting
1550 // struct. To is the result struct built so far, new insertvalue instructions
1552 static Value *BuildSubAggregate(Value *From, Value* To, Type *IndexedType,
1553 SmallVectorImpl<unsigned> &Idxs,
1555 Instruction *InsertBefore) {
1556 llvm::StructType *STy = dyn_cast<llvm::StructType>(IndexedType);
1558 // Save the original To argument so we can modify it
1560 // General case, the type indexed by Idxs is a struct
1561 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
1562 // Process each struct element recursively
1565 To = BuildSubAggregate(From, To, STy->getElementType(i), Idxs, IdxSkip,
1569 // Couldn't find any inserted value for this index? Cleanup
1570 while (PrevTo != OrigTo) {
1571 InsertValueInst* Del = cast<InsertValueInst>(PrevTo);
1572 PrevTo = Del->getAggregateOperand();
1573 Del->eraseFromParent();
1575 // Stop processing elements
1579 // If we successfully found a value for each of our subaggregates
1583 // Base case, the type indexed by SourceIdxs is not a struct, or not all of
1584 // the struct's elements had a value that was inserted directly. In the latter
1585 // case, perhaps we can't determine each of the subelements individually, but
1586 // we might be able to find the complete struct somewhere.
1588 // Find the value that is at that particular spot
1589 Value *V = FindInsertedValue(From, Idxs);
1594 // Insert the value in the new (sub) aggregrate
1595 return llvm::InsertValueInst::Create(To, V, makeArrayRef(Idxs).slice(IdxSkip),
1596 "tmp", InsertBefore);
1599 // This helper takes a nested struct and extracts a part of it (which is again a
1600 // struct) into a new value. For example, given the struct:
1601 // { a, { b, { c, d }, e } }
1602 // and the indices "1, 1" this returns
1605 // It does this by inserting an insertvalue for each element in the resulting
1606 // struct, as opposed to just inserting a single struct. This will only work if
1607 // each of the elements of the substruct are known (ie, inserted into From by an
1608 // insertvalue instruction somewhere).
1610 // All inserted insertvalue instructions are inserted before InsertBefore
1611 static Value *BuildSubAggregate(Value *From, ArrayRef<unsigned> idx_range,
1612 Instruction *InsertBefore) {
1613 assert(InsertBefore && "Must have someplace to insert!");
1614 Type *IndexedType = ExtractValueInst::getIndexedType(From->getType(),
1616 Value *To = UndefValue::get(IndexedType);
1617 SmallVector<unsigned, 10> Idxs(idx_range.begin(), idx_range.end());
1618 unsigned IdxSkip = Idxs.size();
1620 return BuildSubAggregate(From, To, IndexedType, Idxs, IdxSkip, InsertBefore);
1623 /// FindInsertedValue - Given an aggregrate and an sequence of indices, see if
1624 /// the scalar value indexed is already around as a register, for example if it
1625 /// were inserted directly into the aggregrate.
1627 /// If InsertBefore is not null, this function will duplicate (modified)
1628 /// insertvalues when a part of a nested struct is extracted.
1629 Value *llvm::FindInsertedValue(Value *V, ArrayRef<unsigned> idx_range,
1630 Instruction *InsertBefore) {
1631 // Nothing to index? Just return V then (this is useful at the end of our
1633 if (idx_range.empty())
1635 // We have indices, so V should have an indexable type.
1636 assert((V->getType()->isStructTy() || V->getType()->isArrayTy()) &&
1637 "Not looking at a struct or array?");
1638 assert(ExtractValueInst::getIndexedType(V->getType(), idx_range) &&
1639 "Invalid indices for type?");
1641 if (Constant *C = dyn_cast<Constant>(V)) {
1642 C = C->getAggregateElement(idx_range[0]);
1643 if (C == 0) return 0;
1644 return FindInsertedValue(C, idx_range.slice(1), InsertBefore);
1647 if (InsertValueInst *I = dyn_cast<InsertValueInst>(V)) {
1648 // Loop the indices for the insertvalue instruction in parallel with the
1649 // requested indices
1650 const unsigned *req_idx = idx_range.begin();
1651 for (const unsigned *i = I->idx_begin(), *e = I->idx_end();
1652 i != e; ++i, ++req_idx) {
1653 if (req_idx == idx_range.end()) {
1654 // We can't handle this without inserting insertvalues
1658 // The requested index identifies a part of a nested aggregate. Handle
1659 // this specially. For example,
1660 // %A = insertvalue { i32, {i32, i32 } } undef, i32 10, 1, 0
1661 // %B = insertvalue { i32, {i32, i32 } } %A, i32 11, 1, 1
1662 // %C = extractvalue {i32, { i32, i32 } } %B, 1
1663 // This can be changed into
1664 // %A = insertvalue {i32, i32 } undef, i32 10, 0
1665 // %C = insertvalue {i32, i32 } %A, i32 11, 1
1666 // which allows the unused 0,0 element from the nested struct to be
1668 return BuildSubAggregate(V, makeArrayRef(idx_range.begin(), req_idx),
1672 // This insert value inserts something else than what we are looking for.
1673 // See if the (aggregrate) value inserted into has the value we are
1674 // looking for, then.
1676 return FindInsertedValue(I->getAggregateOperand(), idx_range,
1679 // If we end up here, the indices of the insertvalue match with those
1680 // requested (though possibly only partially). Now we recursively look at
1681 // the inserted value, passing any remaining indices.
1682 return FindInsertedValue(I->getInsertedValueOperand(),
1683 makeArrayRef(req_idx, idx_range.end()),
1687 if (ExtractValueInst *I = dyn_cast<ExtractValueInst>(V)) {
1688 // If we're extracting a value from an aggregrate that was extracted from
1689 // something else, we can extract from that something else directly instead.
1690 // However, we will need to chain I's indices with the requested indices.
1692 // Calculate the number of indices required
1693 unsigned size = I->getNumIndices() + idx_range.size();
1694 // Allocate some space to put the new indices in
1695 SmallVector<unsigned, 5> Idxs;
1697 // Add indices from the extract value instruction
1698 Idxs.append(I->idx_begin(), I->idx_end());
1700 // Add requested indices
1701 Idxs.append(idx_range.begin(), idx_range.end());
1703 assert(Idxs.size() == size
1704 && "Number of indices added not correct?");
1706 return FindInsertedValue(I->getAggregateOperand(), Idxs, InsertBefore);
1708 // Otherwise, we don't know (such as, extracting from a function return value
1709 // or load instruction)
1713 /// GetPointerBaseWithConstantOffset - Analyze the specified pointer to see if
1714 /// it can be expressed as a base pointer plus a constant offset. Return the
1715 /// base and offset to the caller.
1716 Value *llvm::GetPointerBaseWithConstantOffset(Value *Ptr, int64_t &Offset,
1717 const DataLayout *DL) {
1718 // Without DataLayout, conservatively assume 64-bit offsets, which is
1719 // the widest we support.
1720 unsigned BitWidth = DL ? DL->getPointerTypeSizeInBits(Ptr->getType()) : 64;
1721 APInt ByteOffset(BitWidth, 0);
1723 if (Ptr->getType()->isVectorTy())
1726 if (GEPOperator *GEP = dyn_cast<GEPOperator>(Ptr)) {
1728 APInt GEPOffset(BitWidth, 0);
1729 if (!GEP->accumulateConstantOffset(*DL, GEPOffset))
1732 ByteOffset += GEPOffset;
1735 Ptr = GEP->getPointerOperand();
1736 } else if (Operator::getOpcode(Ptr) == Instruction::BitCast) {
1737 Ptr = cast<Operator>(Ptr)->getOperand(0);
1738 } else if (GlobalAlias *GA = dyn_cast<GlobalAlias>(Ptr)) {
1739 if (GA->mayBeOverridden())
1741 Ptr = GA->getAliasee();
1746 Offset = ByteOffset.getSExtValue();
1751 /// getConstantStringInfo - This function computes the length of a
1752 /// null-terminated C string pointed to by V. If successful, it returns true
1753 /// and returns the string in Str. If unsuccessful, it returns false.
1754 bool llvm::getConstantStringInfo(const Value *V, StringRef &Str,
1755 uint64_t Offset, bool TrimAtNul) {
1758 // Look through bitcast instructions and geps.
1759 V = V->stripPointerCasts();
1761 // If the value is a GEP instructionor constant expression, treat it as an
1763 if (const GEPOperator *GEP = dyn_cast<GEPOperator>(V)) {
1764 // Make sure the GEP has exactly three arguments.
1765 if (GEP->getNumOperands() != 3)
1768 // Make sure the index-ee is a pointer to array of i8.
1769 PointerType *PT = cast<PointerType>(GEP->getOperand(0)->getType());
1770 ArrayType *AT = dyn_cast<ArrayType>(PT->getElementType());
1771 if (AT == 0 || !AT->getElementType()->isIntegerTy(8))
1774 // Check to make sure that the first operand of the GEP is an integer and
1775 // has value 0 so that we are sure we're indexing into the initializer.
1776 const ConstantInt *FirstIdx = dyn_cast<ConstantInt>(GEP->getOperand(1));
1777 if (FirstIdx == 0 || !FirstIdx->isZero())
1780 // If the second index isn't a ConstantInt, then this is a variable index
1781 // into the array. If this occurs, we can't say anything meaningful about
1783 uint64_t StartIdx = 0;
1784 if (const ConstantInt *CI = dyn_cast<ConstantInt>(GEP->getOperand(2)))
1785 StartIdx = CI->getZExtValue();
1788 return getConstantStringInfo(GEP->getOperand(0), Str, StartIdx+Offset);
1791 // The GEP instruction, constant or instruction, must reference a global
1792 // variable that is a constant and is initialized. The referenced constant
1793 // initializer is the array that we'll use for optimization.
1794 const GlobalVariable *GV = dyn_cast<GlobalVariable>(V);
1795 if (!GV || !GV->isConstant() || !GV->hasDefinitiveInitializer())
1798 // Handle the all-zeros case
1799 if (GV->getInitializer()->isNullValue()) {
1800 // This is a degenerate case. The initializer is constant zero so the
1801 // length of the string must be zero.
1806 // Must be a Constant Array
1807 const ConstantDataArray *Array =
1808 dyn_cast<ConstantDataArray>(GV->getInitializer());
1809 if (Array == 0 || !Array->isString())
1812 // Get the number of elements in the array
1813 uint64_t NumElts = Array->getType()->getArrayNumElements();
1815 // Start out with the entire array in the StringRef.
1816 Str = Array->getAsString();
1818 if (Offset > NumElts)
1821 // Skip over 'offset' bytes.
1822 Str = Str.substr(Offset);
1825 // Trim off the \0 and anything after it. If the array is not nul
1826 // terminated, we just return the whole end of string. The client may know
1827 // some other way that the string is length-bound.
1828 Str = Str.substr(0, Str.find('\0'));
1833 // These next two are very similar to the above, but also look through PHI
1835 // TODO: See if we can integrate these two together.
1837 /// GetStringLengthH - If we can compute the length of the string pointed to by
1838 /// the specified pointer, return 'len+1'. If we can't, return 0.
1839 static uint64_t GetStringLengthH(Value *V, SmallPtrSet<PHINode*, 32> &PHIs) {
1840 // Look through noop bitcast instructions.
1841 V = V->stripPointerCasts();
1843 // If this is a PHI node, there are two cases: either we have already seen it
1845 if (PHINode *PN = dyn_cast<PHINode>(V)) {
1846 if (!PHIs.insert(PN))
1847 return ~0ULL; // already in the set.
1849 // If it was new, see if all the input strings are the same length.
1850 uint64_t LenSoFar = ~0ULL;
1851 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
1852 uint64_t Len = GetStringLengthH(PN->getIncomingValue(i), PHIs);
1853 if (Len == 0) return 0; // Unknown length -> unknown.
1855 if (Len == ~0ULL) continue;
1857 if (Len != LenSoFar && LenSoFar != ~0ULL)
1858 return 0; // Disagree -> unknown.
1862 // Success, all agree.
1866 // strlen(select(c,x,y)) -> strlen(x) ^ strlen(y)
1867 if (SelectInst *SI = dyn_cast<SelectInst>(V)) {
1868 uint64_t Len1 = GetStringLengthH(SI->getTrueValue(), PHIs);
1869 if (Len1 == 0) return 0;
1870 uint64_t Len2 = GetStringLengthH(SI->getFalseValue(), PHIs);
1871 if (Len2 == 0) return 0;
1872 if (Len1 == ~0ULL) return Len2;
1873 if (Len2 == ~0ULL) return Len1;
1874 if (Len1 != Len2) return 0;
1878 // Otherwise, see if we can read the string.
1880 if (!getConstantStringInfo(V, StrData))
1883 return StrData.size()+1;
1886 /// GetStringLength - If we can compute the length of the string pointed to by
1887 /// the specified pointer, return 'len+1'. If we can't, return 0.
1888 uint64_t llvm::GetStringLength(Value *V) {
1889 if (!V->getType()->isPointerTy()) return 0;
1891 SmallPtrSet<PHINode*, 32> PHIs;
1892 uint64_t Len = GetStringLengthH(V, PHIs);
1893 // If Len is ~0ULL, we had an infinite phi cycle: this is dead code, so return
1894 // an empty string as a length.
1895 return Len == ~0ULL ? 1 : Len;
1899 llvm::GetUnderlyingObject(Value *V, const DataLayout *TD, unsigned MaxLookup) {
1900 if (!V->getType()->isPointerTy())
1902 for (unsigned Count = 0; MaxLookup == 0 || Count < MaxLookup; ++Count) {
1903 if (GEPOperator *GEP = dyn_cast<GEPOperator>(V)) {
1904 V = GEP->getPointerOperand();
1905 } else if (Operator::getOpcode(V) == Instruction::BitCast) {
1906 V = cast<Operator>(V)->getOperand(0);
1907 } else if (GlobalAlias *GA = dyn_cast<GlobalAlias>(V)) {
1908 if (GA->mayBeOverridden())
1910 V = GA->getAliasee();
1912 // See if InstructionSimplify knows any relevant tricks.
1913 if (Instruction *I = dyn_cast<Instruction>(V))
1914 // TODO: Acquire a DominatorTree and use it.
1915 if (Value *Simplified = SimplifyInstruction(I, TD, 0)) {
1922 assert(V->getType()->isPointerTy() && "Unexpected operand type!");
1928 llvm::GetUnderlyingObjects(Value *V,
1929 SmallVectorImpl<Value *> &Objects,
1930 const DataLayout *TD,
1931 unsigned MaxLookup) {
1932 SmallPtrSet<Value *, 4> Visited;
1933 SmallVector<Value *, 4> Worklist;
1934 Worklist.push_back(V);
1936 Value *P = Worklist.pop_back_val();
1937 P = GetUnderlyingObject(P, TD, MaxLookup);
1939 if (!Visited.insert(P))
1942 if (SelectInst *SI = dyn_cast<SelectInst>(P)) {
1943 Worklist.push_back(SI->getTrueValue());
1944 Worklist.push_back(SI->getFalseValue());
1948 if (PHINode *PN = dyn_cast<PHINode>(P)) {
1949 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
1950 Worklist.push_back(PN->getIncomingValue(i));
1954 Objects.push_back(P);
1955 } while (!Worklist.empty());
1958 /// onlyUsedByLifetimeMarkers - Return true if the only users of this pointer
1959 /// are lifetime markers.
1961 bool llvm::onlyUsedByLifetimeMarkers(const Value *V) {
1962 for (Value::const_use_iterator UI = V->use_begin(), UE = V->use_end();
1964 const IntrinsicInst *II = dyn_cast<IntrinsicInst>(*UI);
1965 if (!II) return false;
1967 if (II->getIntrinsicID() != Intrinsic::lifetime_start &&
1968 II->getIntrinsicID() != Intrinsic::lifetime_end)
1974 bool llvm::isSafeToSpeculativelyExecute(const Value *V,
1975 const DataLayout *TD) {
1976 const Operator *Inst = dyn_cast<Operator>(V);
1980 for (unsigned i = 0, e = Inst->getNumOperands(); i != e; ++i)
1981 if (Constant *C = dyn_cast<Constant>(Inst->getOperand(i)))
1985 switch (Inst->getOpcode()) {
1988 case Instruction::UDiv:
1989 case Instruction::URem:
1990 // x / y is undefined if y == 0, but calcuations like x / 3 are safe.
1991 return isKnownNonZero(Inst->getOperand(1), TD);
1992 case Instruction::SDiv:
1993 case Instruction::SRem: {
1994 Value *Op = Inst->getOperand(1);
1995 // x / y is undefined if y == 0
1996 if (!isKnownNonZero(Op, TD))
1998 // x / y might be undefined if y == -1
1999 unsigned BitWidth = getBitWidth(Op->getType(), TD);
2002 APInt KnownZero(BitWidth, 0);
2003 APInt KnownOne(BitWidth, 0);
2004 ComputeMaskedBits(Op, KnownZero, KnownOne, TD);
2007 case Instruction::Load: {
2008 const LoadInst *LI = cast<LoadInst>(Inst);
2009 if (!LI->isUnordered())
2011 return LI->getPointerOperand()->isDereferenceablePointer();
2013 case Instruction::Call: {
2014 if (const IntrinsicInst *II = dyn_cast<IntrinsicInst>(Inst)) {
2015 switch (II->getIntrinsicID()) {
2016 // These synthetic intrinsics have no side-effects, and just mark
2017 // information about their operands.
2018 // FIXME: There are other no-op synthetic instructions that potentially
2019 // should be considered at least *safe* to speculate...
2020 case Intrinsic::dbg_declare:
2021 case Intrinsic::dbg_value:
2024 case Intrinsic::bswap:
2025 case Intrinsic::ctlz:
2026 case Intrinsic::ctpop:
2027 case Intrinsic::cttz:
2028 case Intrinsic::objectsize:
2029 case Intrinsic::sadd_with_overflow:
2030 case Intrinsic::smul_with_overflow:
2031 case Intrinsic::ssub_with_overflow:
2032 case Intrinsic::uadd_with_overflow:
2033 case Intrinsic::umul_with_overflow:
2034 case Intrinsic::usub_with_overflow:
2036 // TODO: some fp intrinsics are marked as having the same error handling
2037 // as libm. They're safe to speculate when they won't error.
2038 // TODO: are convert_{from,to}_fp16 safe?
2039 // TODO: can we list target-specific intrinsics here?
2043 return false; // The called function could have undefined behavior or
2044 // side-effects, even if marked readnone nounwind.
2046 case Instruction::VAArg:
2047 case Instruction::Alloca:
2048 case Instruction::Invoke:
2049 case Instruction::PHI:
2050 case Instruction::Store:
2051 case Instruction::Ret:
2052 case Instruction::Br:
2053 case Instruction::IndirectBr:
2054 case Instruction::Switch:
2055 case Instruction::Unreachable:
2056 case Instruction::Fence:
2057 case Instruction::LandingPad:
2058 case Instruction::AtomicRMW:
2059 case Instruction::AtomicCmpXchg:
2060 case Instruction::Resume:
2061 return false; // Misc instructions which have effects
2065 /// isKnownNonNull - Return true if we know that the specified value is never
2067 bool llvm::isKnownNonNull(const Value *V, const TargetLibraryInfo *TLI) {
2068 // Alloca never returns null, malloc might.
2069 if (isa<AllocaInst>(V)) return true;
2071 // A byval argument is never null.
2072 if (const Argument *A = dyn_cast<Argument>(V))
2073 return A->hasByValAttr();
2075 // Global values are not null unless extern weak.
2076 if (const GlobalValue *GV = dyn_cast<GlobalValue>(V))
2077 return !GV->hasExternalWeakLinkage();
2079 // operator new never returns null.
2080 if (isOperatorNewLikeFn(V, TLI, /*LookThroughBitCast=*/true))