1 //===- ScalarEvolution.cpp - Scalar Evolution Analysis ----------*- C++ -*-===//
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 the implementation of the scalar evolution analysis
11 // engine, which is used primarily to analyze expressions involving induction
12 // variables in loops.
14 // There are several aspects to this library. First is the representation of
15 // scalar expressions, which are represented as subclasses of the SCEV class.
16 // These classes are used to represent certain types of subexpressions that we
17 // can handle. These classes are reference counted, managed by the SCEVHandle
18 // class. We only create one SCEV of a particular shape, so pointer-comparisons
19 // for equality are legal.
21 // One important aspect of the SCEV objects is that they are never cyclic, even
22 // if there is a cycle in the dataflow for an expression (ie, a PHI node). If
23 // the PHI node is one of the idioms that we can represent (e.g., a polynomial
24 // recurrence) then we represent it directly as a recurrence node, otherwise we
25 // represent it as a SCEVUnknown node.
27 // In addition to being able to represent expressions of various types, we also
28 // have folders that are used to build the *canonical* representation for a
29 // particular expression. These folders are capable of using a variety of
30 // rewrite rules to simplify the expressions.
32 // Once the folders are defined, we can implement the more interesting
33 // higher-level code, such as the code that recognizes PHI nodes of various
34 // types, computes the execution count of a loop, etc.
36 // TODO: We should use these routines and value representations to implement
37 // dependence analysis!
39 //===----------------------------------------------------------------------===//
41 // There are several good references for the techniques used in this analysis.
43 // Chains of recurrences -- a method to expedite the evaluation
44 // of closed-form functions
45 // Olaf Bachmann, Paul S. Wang, Eugene V. Zima
47 // On computational properties of chains of recurrences
50 // Symbolic Evaluation of Chains of Recurrences for Loop Optimization
51 // Robert A. van Engelen
53 // Efficient Symbolic Analysis for Optimizing Compilers
54 // Robert A. van Engelen
56 // Using the chains of recurrences algebra for data dependence testing and
57 // induction variable substitution
58 // MS Thesis, Johnie Birch
60 //===----------------------------------------------------------------------===//
62 #define DEBUG_TYPE "scalar-evolution"
63 #include "llvm/Analysis/ScalarEvolutionExpressions.h"
64 #include "llvm/Constants.h"
65 #include "llvm/DerivedTypes.h"
66 #include "llvm/GlobalVariable.h"
67 #include "llvm/Instructions.h"
68 #include "llvm/Analysis/ConstantFolding.h"
69 #include "llvm/Analysis/LoopInfo.h"
70 #include "llvm/Assembly/Writer.h"
71 #include "llvm/Transforms/Scalar.h"
72 #include "llvm/Support/CFG.h"
73 #include "llvm/Support/CommandLine.h"
74 #include "llvm/Support/Compiler.h"
75 #include "llvm/Support/ConstantRange.h"
76 #include "llvm/Support/InstIterator.h"
77 #include "llvm/Support/ManagedStatic.h"
78 #include "llvm/Support/MathExtras.h"
79 #include "llvm/Support/Streams.h"
80 #include "llvm/ADT/Statistic.h"
86 STATISTIC(NumBruteForceEvaluations,
87 "Number of brute force evaluations needed to "
88 "calculate high-order polynomial exit values");
89 STATISTIC(NumArrayLenItCounts,
90 "Number of trip counts computed with array length");
91 STATISTIC(NumTripCountsComputed,
92 "Number of loops with predictable loop counts");
93 STATISTIC(NumTripCountsNotComputed,
94 "Number of loops without predictable loop counts");
95 STATISTIC(NumBruteForceTripCountsComputed,
96 "Number of loops with trip counts computed by force");
98 static cl::opt<unsigned>
99 MaxBruteForceIterations("scalar-evolution-max-iterations", cl::ReallyHidden,
100 cl::desc("Maximum number of iterations SCEV will "
101 "symbolically execute a constant derived loop"),
104 static RegisterPass<ScalarEvolution>
105 R("scalar-evolution", "Scalar Evolution Analysis", false, true);
106 char ScalarEvolution::ID = 0;
108 //===----------------------------------------------------------------------===//
109 // SCEV class definitions
110 //===----------------------------------------------------------------------===//
112 //===----------------------------------------------------------------------===//
113 // Implementation of the SCEV class.
116 void SCEV::dump() const {
120 uint32_t SCEV::getBitWidth() const {
121 if (const IntegerType* ITy = dyn_cast<IntegerType>(getType()))
122 return ITy->getBitWidth();
126 bool SCEV::isZero() const {
127 if (const SCEVConstant *SC = dyn_cast<SCEVConstant>(this))
128 return SC->getValue()->isZero();
133 SCEVCouldNotCompute::SCEVCouldNotCompute() : SCEV(scCouldNotCompute) {}
135 bool SCEVCouldNotCompute::isLoopInvariant(const Loop *L) const {
136 assert(0 && "Attempt to use a SCEVCouldNotCompute object!");
140 const Type *SCEVCouldNotCompute::getType() const {
141 assert(0 && "Attempt to use a SCEVCouldNotCompute object!");
145 bool SCEVCouldNotCompute::hasComputableLoopEvolution(const Loop *L) const {
146 assert(0 && "Attempt to use a SCEVCouldNotCompute object!");
150 SCEVHandle SCEVCouldNotCompute::
151 replaceSymbolicValuesWithConcrete(const SCEVHandle &Sym,
152 const SCEVHandle &Conc,
153 ScalarEvolution &SE) const {
157 void SCEVCouldNotCompute::print(std::ostream &OS) const {
158 OS << "***COULDNOTCOMPUTE***";
161 bool SCEVCouldNotCompute::classof(const SCEV *S) {
162 return S->getSCEVType() == scCouldNotCompute;
166 // SCEVConstants - Only allow the creation of one SCEVConstant for any
167 // particular value. Don't use a SCEVHandle here, or else the object will
169 static ManagedStatic<std::map<ConstantInt*, SCEVConstant*> > SCEVConstants;
172 SCEVConstant::~SCEVConstant() {
173 SCEVConstants->erase(V);
176 SCEVHandle ScalarEvolution::getConstant(ConstantInt *V) {
177 SCEVConstant *&R = (*SCEVConstants)[V];
178 if (R == 0) R = new SCEVConstant(V);
182 SCEVHandle ScalarEvolution::getConstant(const APInt& Val) {
183 return getConstant(ConstantInt::get(Val));
186 const Type *SCEVConstant::getType() const { return V->getType(); }
188 void SCEVConstant::print(std::ostream &OS) const {
189 WriteAsOperand(OS, V, false);
192 // SCEVTruncates - Only allow the creation of one SCEVTruncateExpr for any
193 // particular input. Don't use a SCEVHandle here, or else the object will
195 static ManagedStatic<std::map<std::pair<SCEV*, const Type*>,
196 SCEVTruncateExpr*> > SCEVTruncates;
198 SCEVTruncateExpr::SCEVTruncateExpr(const SCEVHandle &op, const Type *ty)
199 : SCEV(scTruncate), Op(op), Ty(ty) {
200 assert(Op->getType()->isInteger() && Ty->isInteger() &&
201 "Cannot truncate non-integer value!");
202 assert(Op->getType()->getPrimitiveSizeInBits() > Ty->getPrimitiveSizeInBits()
203 && "This is not a truncating conversion!");
206 SCEVTruncateExpr::~SCEVTruncateExpr() {
207 SCEVTruncates->erase(std::make_pair(Op, Ty));
210 void SCEVTruncateExpr::print(std::ostream &OS) const {
211 OS << "(truncate " << *Op << " to " << *Ty << ")";
214 // SCEVZeroExtends - Only allow the creation of one SCEVZeroExtendExpr for any
215 // particular input. Don't use a SCEVHandle here, or else the object will never
217 static ManagedStatic<std::map<std::pair<SCEV*, const Type*>,
218 SCEVZeroExtendExpr*> > SCEVZeroExtends;
220 SCEVZeroExtendExpr::SCEVZeroExtendExpr(const SCEVHandle &op, const Type *ty)
221 : SCEV(scZeroExtend), Op(op), Ty(ty) {
222 assert(Op->getType()->isInteger() && Ty->isInteger() &&
223 "Cannot zero extend non-integer value!");
224 assert(Op->getType()->getPrimitiveSizeInBits() < Ty->getPrimitiveSizeInBits()
225 && "This is not an extending conversion!");
228 SCEVZeroExtendExpr::~SCEVZeroExtendExpr() {
229 SCEVZeroExtends->erase(std::make_pair(Op, Ty));
232 void SCEVZeroExtendExpr::print(std::ostream &OS) const {
233 OS << "(zeroextend " << *Op << " to " << *Ty << ")";
236 // SCEVSignExtends - Only allow the creation of one SCEVSignExtendExpr for any
237 // particular input. Don't use a SCEVHandle here, or else the object will never
239 static ManagedStatic<std::map<std::pair<SCEV*, const Type*>,
240 SCEVSignExtendExpr*> > SCEVSignExtends;
242 SCEVSignExtendExpr::SCEVSignExtendExpr(const SCEVHandle &op, const Type *ty)
243 : SCEV(scSignExtend), Op(op), Ty(ty) {
244 assert(Op->getType()->isInteger() && Ty->isInteger() &&
245 "Cannot sign extend non-integer value!");
246 assert(Op->getType()->getPrimitiveSizeInBits() < Ty->getPrimitiveSizeInBits()
247 && "This is not an extending conversion!");
250 SCEVSignExtendExpr::~SCEVSignExtendExpr() {
251 SCEVSignExtends->erase(std::make_pair(Op, Ty));
254 void SCEVSignExtendExpr::print(std::ostream &OS) const {
255 OS << "(signextend " << *Op << " to " << *Ty << ")";
258 // SCEVCommExprs - Only allow the creation of one SCEVCommutativeExpr for any
259 // particular input. Don't use a SCEVHandle here, or else the object will never
261 static ManagedStatic<std::map<std::pair<unsigned, std::vector<SCEV*> >,
262 SCEVCommutativeExpr*> > SCEVCommExprs;
264 SCEVCommutativeExpr::~SCEVCommutativeExpr() {
265 SCEVCommExprs->erase(std::make_pair(getSCEVType(),
266 std::vector<SCEV*>(Operands.begin(),
270 void SCEVCommutativeExpr::print(std::ostream &OS) const {
271 assert(Operands.size() > 1 && "This plus expr shouldn't exist!");
272 const char *OpStr = getOperationStr();
273 OS << "(" << *Operands[0];
274 for (unsigned i = 1, e = Operands.size(); i != e; ++i)
275 OS << OpStr << *Operands[i];
279 SCEVHandle SCEVCommutativeExpr::
280 replaceSymbolicValuesWithConcrete(const SCEVHandle &Sym,
281 const SCEVHandle &Conc,
282 ScalarEvolution &SE) const {
283 for (unsigned i = 0, e = getNumOperands(); i != e; ++i) {
285 getOperand(i)->replaceSymbolicValuesWithConcrete(Sym, Conc, SE);
286 if (H != getOperand(i)) {
287 std::vector<SCEVHandle> NewOps;
288 NewOps.reserve(getNumOperands());
289 for (unsigned j = 0; j != i; ++j)
290 NewOps.push_back(getOperand(j));
292 for (++i; i != e; ++i)
293 NewOps.push_back(getOperand(i)->
294 replaceSymbolicValuesWithConcrete(Sym, Conc, SE));
296 if (isa<SCEVAddExpr>(this))
297 return SE.getAddExpr(NewOps);
298 else if (isa<SCEVMulExpr>(this))
299 return SE.getMulExpr(NewOps);
300 else if (isa<SCEVSMaxExpr>(this))
301 return SE.getSMaxExpr(NewOps);
302 else if (isa<SCEVUMaxExpr>(this))
303 return SE.getUMaxExpr(NewOps);
305 assert(0 && "Unknown commutative expr!");
312 // SCEVUDivs - Only allow the creation of one SCEVUDivExpr for any particular
313 // input. Don't use a SCEVHandle here, or else the object will never be
315 static ManagedStatic<std::map<std::pair<SCEV*, SCEV*>,
316 SCEVUDivExpr*> > SCEVUDivs;
318 SCEVUDivExpr::~SCEVUDivExpr() {
319 SCEVUDivs->erase(std::make_pair(LHS, RHS));
322 void SCEVUDivExpr::print(std::ostream &OS) const {
323 OS << "(" << *LHS << " /u " << *RHS << ")";
326 const Type *SCEVUDivExpr::getType() const {
327 return LHS->getType();
330 // SCEVAddRecExprs - Only allow the creation of one SCEVAddRecExpr for any
331 // particular input. Don't use a SCEVHandle here, or else the object will never
333 static ManagedStatic<std::map<std::pair<const Loop *, std::vector<SCEV*> >,
334 SCEVAddRecExpr*> > SCEVAddRecExprs;
336 SCEVAddRecExpr::~SCEVAddRecExpr() {
337 SCEVAddRecExprs->erase(std::make_pair(L,
338 std::vector<SCEV*>(Operands.begin(),
342 SCEVHandle SCEVAddRecExpr::
343 replaceSymbolicValuesWithConcrete(const SCEVHandle &Sym,
344 const SCEVHandle &Conc,
345 ScalarEvolution &SE) const {
346 for (unsigned i = 0, e = getNumOperands(); i != e; ++i) {
348 getOperand(i)->replaceSymbolicValuesWithConcrete(Sym, Conc, SE);
349 if (H != getOperand(i)) {
350 std::vector<SCEVHandle> NewOps;
351 NewOps.reserve(getNumOperands());
352 for (unsigned j = 0; j != i; ++j)
353 NewOps.push_back(getOperand(j));
355 for (++i; i != e; ++i)
356 NewOps.push_back(getOperand(i)->
357 replaceSymbolicValuesWithConcrete(Sym, Conc, SE));
359 return SE.getAddRecExpr(NewOps, L);
366 bool SCEVAddRecExpr::isLoopInvariant(const Loop *QueryLoop) const {
367 // This recurrence is invariant w.r.t to QueryLoop iff QueryLoop doesn't
368 // contain L and if the start is invariant.
369 return !QueryLoop->contains(L->getHeader()) &&
370 getOperand(0)->isLoopInvariant(QueryLoop);
374 void SCEVAddRecExpr::print(std::ostream &OS) const {
375 OS << "{" << *Operands[0];
376 for (unsigned i = 1, e = Operands.size(); i != e; ++i)
377 OS << ",+," << *Operands[i];
378 OS << "}<" << L->getHeader()->getName() + ">";
381 // SCEVUnknowns - Only allow the creation of one SCEVUnknown for any particular
382 // value. Don't use a SCEVHandle here, or else the object will never be
384 static ManagedStatic<std::map<Value*, SCEVUnknown*> > SCEVUnknowns;
386 SCEVUnknown::~SCEVUnknown() { SCEVUnknowns->erase(V); }
388 bool SCEVUnknown::isLoopInvariant(const Loop *L) const {
389 // All non-instruction values are loop invariant. All instructions are loop
390 // invariant if they are not contained in the specified loop.
391 if (Instruction *I = dyn_cast<Instruction>(V))
392 return !L->contains(I->getParent());
396 const Type *SCEVUnknown::getType() const {
400 void SCEVUnknown::print(std::ostream &OS) const {
401 WriteAsOperand(OS, V, false);
404 //===----------------------------------------------------------------------===//
406 //===----------------------------------------------------------------------===//
409 /// SCEVComplexityCompare - Return true if the complexity of the LHS is less
410 /// than the complexity of the RHS. This comparator is used to canonicalize
412 struct VISIBILITY_HIDDEN SCEVComplexityCompare {
413 bool operator()(const SCEV *LHS, const SCEV *RHS) const {
414 return LHS->getSCEVType() < RHS->getSCEVType();
419 /// GroupByComplexity - Given a list of SCEV objects, order them by their
420 /// complexity, and group objects of the same complexity together by value.
421 /// When this routine is finished, we know that any duplicates in the vector are
422 /// consecutive and that complexity is monotonically increasing.
424 /// Note that we go take special precautions to ensure that we get determinstic
425 /// results from this routine. In other words, we don't want the results of
426 /// this to depend on where the addresses of various SCEV objects happened to
429 static void GroupByComplexity(std::vector<SCEVHandle> &Ops) {
430 if (Ops.size() < 2) return; // Noop
431 if (Ops.size() == 2) {
432 // This is the common case, which also happens to be trivially simple.
434 if (SCEVComplexityCompare()(Ops[1], Ops[0]))
435 std::swap(Ops[0], Ops[1]);
439 // Do the rough sort by complexity.
440 std::sort(Ops.begin(), Ops.end(), SCEVComplexityCompare());
442 // Now that we are sorted by complexity, group elements of the same
443 // complexity. Note that this is, at worst, N^2, but the vector is likely to
444 // be extremely short in practice. Note that we take this approach because we
445 // do not want to depend on the addresses of the objects we are grouping.
446 for (unsigned i = 0, e = Ops.size(); i != e-2; ++i) {
448 unsigned Complexity = S->getSCEVType();
450 // If there are any objects of the same complexity and same value as this
452 for (unsigned j = i+1; j != e && Ops[j]->getSCEVType() == Complexity; ++j) {
453 if (Ops[j] == S) { // Found a duplicate.
454 // Move it to immediately after i'th element.
455 std::swap(Ops[i+1], Ops[j]);
456 ++i; // no need to rescan it.
457 if (i == e-2) return; // Done!
465 //===----------------------------------------------------------------------===//
466 // Simple SCEV method implementations
467 //===----------------------------------------------------------------------===//
469 /// getIntegerSCEV - Given an integer or FP type, create a constant for the
470 /// specified signed integer value and return a SCEV for the constant.
471 SCEVHandle ScalarEvolution::getIntegerSCEV(int Val, const Type *Ty) {
474 C = Constant::getNullValue(Ty);
475 else if (Ty->isFloatingPoint())
476 C = ConstantFP::get(APFloat(Ty==Type::FloatTy ? APFloat::IEEEsingle :
477 APFloat::IEEEdouble, Val));
479 C = ConstantInt::get(Ty, Val);
480 return getUnknown(C);
483 /// getNegativeSCEV - Return a SCEV corresponding to -V = -1*V
485 SCEVHandle ScalarEvolution::getNegativeSCEV(const SCEVHandle &V) {
486 if (SCEVConstant *VC = dyn_cast<SCEVConstant>(V))
487 return getUnknown(ConstantExpr::getNeg(VC->getValue()));
489 return getMulExpr(V, getConstant(ConstantInt::getAllOnesValue(V->getType())));
492 /// getNotSCEV - Return a SCEV corresponding to ~V = -1-V
493 SCEVHandle ScalarEvolution::getNotSCEV(const SCEVHandle &V) {
494 if (SCEVConstant *VC = dyn_cast<SCEVConstant>(V))
495 return getUnknown(ConstantExpr::getNot(VC->getValue()));
497 SCEVHandle AllOnes = getConstant(ConstantInt::getAllOnesValue(V->getType()));
498 return getMinusSCEV(AllOnes, V);
501 /// getMinusSCEV - Return a SCEV corresponding to LHS - RHS.
503 SCEVHandle ScalarEvolution::getMinusSCEV(const SCEVHandle &LHS,
504 const SCEVHandle &RHS) {
506 return getAddExpr(LHS, getNegativeSCEV(RHS));
510 /// BinomialCoefficient - Compute BC(It, K). The result is of the same type as
511 /// It. Assume, K > 0.
512 static SCEVHandle BinomialCoefficient(SCEVHandle It, unsigned K,
513 ScalarEvolution &SE) {
514 // We are using the following formula for BC(It, K):
516 // BC(It, K) = (It * (It - 1) * ... * (It - K + 1)) / K!
518 // Suppose, W is the bitwidth of It (and of the return value as well). We
519 // must be prepared for overflow. Hence, we must assure that the result of
520 // our computation is equal to the accurate one modulo 2^W. Unfortunately,
521 // division isn't safe in modular arithmetic. This means we must perform the
522 // whole computation accurately and then truncate the result to W bits.
524 // The dividend of the formula is a multiplication of K integers of bitwidth
525 // W. K*W bits suffice to compute it accurately.
527 // FIXME: We assume the divisor can be accurately computed using 16-bit
528 // unsigned integer type. It is true up to K = 8 (AddRecs of length 9). In
529 // future we may use APInt to use the minimum number of bits necessary to
530 // compute it accurately.
532 // It is safe to use unsigned division here: the dividend is nonnegative and
533 // the divisor is positive.
535 // Handle the simplest case efficiently.
539 assert(K < 9 && "We cannot handle such long AddRecs yet.");
541 unsigned DividendBits = K * It->getBitWidth();
542 if (DividendBits > 256)
543 return new SCEVCouldNotCompute();
545 const IntegerType *DividendTy = IntegerType::get(DividendBits);
546 const SCEVHandle ExIt = SE.getZeroExtendExpr(It, DividendTy);
548 // The final number of bits we need to perform the division is the maximum of
549 // dividend and divisor bitwidths.
550 const IntegerType *DivisionTy =
551 IntegerType::get(std::max(DividendBits, 16U));
553 // Compute K! We know K >= 2 here.
555 for (unsigned i = 3; i <= K; ++i)
557 APInt Divisor(DivisionTy->getBitWidth(), F);
559 // Handle this case efficiently, it is common to have constant iteration
560 // counts while computing loop exit values.
561 if (SCEVConstant *SC = dyn_cast<SCEVConstant>(ExIt)) {
562 const APInt& N = SC->getValue()->getValue();
563 APInt Dividend(N.getBitWidth(), 1);
566 if (DividendTy != DivisionTy)
567 Dividend = Dividend.zext(DivisionTy->getBitWidth());
568 return SE.getConstant(Dividend.udiv(Divisor).trunc(It->getBitWidth()));
571 SCEVHandle Dividend = ExIt;
572 for (unsigned i = 1; i != K; ++i)
574 SE.getMulExpr(Dividend,
575 SE.getMinusSCEV(ExIt, SE.getIntegerSCEV(i, DividendTy)));
577 if (DividendTy != DivisionTy)
578 Dividend = SE.getZeroExtendExpr(Dividend, DivisionTy);
579 return SE.getTruncateExpr(SE.getUDivExpr(Dividend, SE.getConstant(Divisor)),
583 /// evaluateAtIteration - Return the value of this chain of recurrences at
584 /// the specified iteration number. We can evaluate this recurrence by
585 /// multiplying each element in the chain by the binomial coefficient
586 /// corresponding to it. In other words, we can evaluate {A,+,B,+,C,+,D} as:
588 /// A*BC(It, 0) + B*BC(It, 1) + C*BC(It, 2) + D*BC(It, 3)
590 /// where BC(It, k) stands for binomial coefficient.
592 SCEVHandle SCEVAddRecExpr::evaluateAtIteration(SCEVHandle It,
593 ScalarEvolution &SE) const {
594 SCEVHandle Result = getStart();
595 for (unsigned i = 1, e = getNumOperands(); i != e; ++i) {
596 // The computation is correct in the face of overflow provided that the
597 // multiplication is performed _after_ the evaluation of the binomial
599 SCEVHandle Val = SE.getMulExpr(getOperand(i),
600 BinomialCoefficient(It, i, SE));
601 Result = SE.getAddExpr(Result, Val);
606 //===----------------------------------------------------------------------===//
607 // SCEV Expression folder implementations
608 //===----------------------------------------------------------------------===//
610 SCEVHandle ScalarEvolution::getTruncateExpr(const SCEVHandle &Op, const Type *Ty) {
611 if (SCEVConstant *SC = dyn_cast<SCEVConstant>(Op))
613 ConstantExpr::getTrunc(SC->getValue(), Ty));
615 // If the input value is a chrec scev made out of constants, truncate
616 // all of the constants.
617 if (SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(Op)) {
618 std::vector<SCEVHandle> Operands;
619 for (unsigned i = 0, e = AddRec->getNumOperands(); i != e; ++i)
620 // FIXME: This should allow truncation of other expression types!
621 if (isa<SCEVConstant>(AddRec->getOperand(i)))
622 Operands.push_back(getTruncateExpr(AddRec->getOperand(i), Ty));
625 if (Operands.size() == AddRec->getNumOperands())
626 return getAddRecExpr(Operands, AddRec->getLoop());
629 SCEVTruncateExpr *&Result = (*SCEVTruncates)[std::make_pair(Op, Ty)];
630 if (Result == 0) Result = new SCEVTruncateExpr(Op, Ty);
634 SCEVHandle ScalarEvolution::getZeroExtendExpr(const SCEVHandle &Op, const Type *Ty) {
635 if (SCEVConstant *SC = dyn_cast<SCEVConstant>(Op))
637 ConstantExpr::getZExt(SC->getValue(), Ty));
639 // FIXME: If the input value is a chrec scev, and we can prove that the value
640 // did not overflow the old, smaller, value, we can zero extend all of the
641 // operands (often constants). This would allow analysis of something like
642 // this: for (unsigned char X = 0; X < 100; ++X) { int Y = X; }
644 SCEVZeroExtendExpr *&Result = (*SCEVZeroExtends)[std::make_pair(Op, Ty)];
645 if (Result == 0) Result = new SCEVZeroExtendExpr(Op, Ty);
649 SCEVHandle ScalarEvolution::getSignExtendExpr(const SCEVHandle &Op, const Type *Ty) {
650 if (SCEVConstant *SC = dyn_cast<SCEVConstant>(Op))
652 ConstantExpr::getSExt(SC->getValue(), Ty));
654 // FIXME: If the input value is a chrec scev, and we can prove that the value
655 // did not overflow the old, smaller, value, we can sign extend all of the
656 // operands (often constants). This would allow analysis of something like
657 // this: for (signed char X = 0; X < 100; ++X) { int Y = X; }
659 SCEVSignExtendExpr *&Result = (*SCEVSignExtends)[std::make_pair(Op, Ty)];
660 if (Result == 0) Result = new SCEVSignExtendExpr(Op, Ty);
664 /// getTruncateOrZeroExtend - Return a SCEV corresponding to a conversion
665 /// of the input value to the specified type. If the type must be
666 /// extended, it is zero extended.
667 SCEVHandle ScalarEvolution::getTruncateOrZeroExtend(const SCEVHandle &V,
669 const Type *SrcTy = V->getType();
670 assert(SrcTy->isInteger() && Ty->isInteger() &&
671 "Cannot truncate or zero extend with non-integer arguments!");
672 if (SrcTy->getPrimitiveSizeInBits() == Ty->getPrimitiveSizeInBits())
673 return V; // No conversion
674 if (SrcTy->getPrimitiveSizeInBits() > Ty->getPrimitiveSizeInBits())
675 return getTruncateExpr(V, Ty);
676 return getZeroExtendExpr(V, Ty);
679 // get - Get a canonical add expression, or something simpler if possible.
680 SCEVHandle ScalarEvolution::getAddExpr(std::vector<SCEVHandle> &Ops) {
681 assert(!Ops.empty() && "Cannot get empty add!");
682 if (Ops.size() == 1) return Ops[0];
684 // Sort by complexity, this groups all similar expression types together.
685 GroupByComplexity(Ops);
687 // If there are any constants, fold them together.
689 if (SCEVConstant *LHSC = dyn_cast<SCEVConstant>(Ops[0])) {
691 assert(Idx < Ops.size());
692 while (SCEVConstant *RHSC = dyn_cast<SCEVConstant>(Ops[Idx])) {
693 // We found two constants, fold them together!
694 ConstantInt *Fold = ConstantInt::get(LHSC->getValue()->getValue() +
695 RHSC->getValue()->getValue());
696 Ops[0] = getConstant(Fold);
697 Ops.erase(Ops.begin()+1); // Erase the folded element
698 if (Ops.size() == 1) return Ops[0];
699 LHSC = cast<SCEVConstant>(Ops[0]);
702 // If we are left with a constant zero being added, strip it off.
703 if (cast<SCEVConstant>(Ops[0])->getValue()->isZero()) {
704 Ops.erase(Ops.begin());
709 if (Ops.size() == 1) return Ops[0];
711 // Okay, check to see if the same value occurs in the operand list twice. If
712 // so, merge them together into an multiply expression. Since we sorted the
713 // list, these values are required to be adjacent.
714 const Type *Ty = Ops[0]->getType();
715 for (unsigned i = 0, e = Ops.size()-1; i != e; ++i)
716 if (Ops[i] == Ops[i+1]) { // X + Y + Y --> X + Y*2
717 // Found a match, merge the two values into a multiply, and add any
718 // remaining values to the result.
719 SCEVHandle Two = getIntegerSCEV(2, Ty);
720 SCEVHandle Mul = getMulExpr(Ops[i], Two);
723 Ops.erase(Ops.begin()+i, Ops.begin()+i+2);
725 return getAddExpr(Ops);
728 // Now we know the first non-constant operand. Skip past any cast SCEVs.
729 while (Idx < Ops.size() && Ops[Idx]->getSCEVType() < scAddExpr)
732 // If there are add operands they would be next.
733 if (Idx < Ops.size()) {
734 bool DeletedAdd = false;
735 while (SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(Ops[Idx])) {
736 // If we have an add, expand the add operands onto the end of the operands
738 Ops.insert(Ops.end(), Add->op_begin(), Add->op_end());
739 Ops.erase(Ops.begin()+Idx);
743 // If we deleted at least one add, we added operands to the end of the list,
744 // and they are not necessarily sorted. Recurse to resort and resimplify
745 // any operands we just aquired.
747 return getAddExpr(Ops);
750 // Skip over the add expression until we get to a multiply.
751 while (Idx < Ops.size() && Ops[Idx]->getSCEVType() < scMulExpr)
754 // If we are adding something to a multiply expression, make sure the
755 // something is not already an operand of the multiply. If so, merge it into
757 for (; Idx < Ops.size() && isa<SCEVMulExpr>(Ops[Idx]); ++Idx) {
758 SCEVMulExpr *Mul = cast<SCEVMulExpr>(Ops[Idx]);
759 for (unsigned MulOp = 0, e = Mul->getNumOperands(); MulOp != e; ++MulOp) {
760 SCEV *MulOpSCEV = Mul->getOperand(MulOp);
761 for (unsigned AddOp = 0, e = Ops.size(); AddOp != e; ++AddOp)
762 if (MulOpSCEV == Ops[AddOp] && !isa<SCEVConstant>(MulOpSCEV)) {
763 // Fold W + X + (X * Y * Z) --> W + (X * ((Y*Z)+1))
764 SCEVHandle InnerMul = Mul->getOperand(MulOp == 0);
765 if (Mul->getNumOperands() != 2) {
766 // If the multiply has more than two operands, we must get the
768 std::vector<SCEVHandle> MulOps(Mul->op_begin(), Mul->op_end());
769 MulOps.erase(MulOps.begin()+MulOp);
770 InnerMul = getMulExpr(MulOps);
772 SCEVHandle One = getIntegerSCEV(1, Ty);
773 SCEVHandle AddOne = getAddExpr(InnerMul, One);
774 SCEVHandle OuterMul = getMulExpr(AddOne, Ops[AddOp]);
775 if (Ops.size() == 2) return OuterMul;
777 Ops.erase(Ops.begin()+AddOp);
778 Ops.erase(Ops.begin()+Idx-1);
780 Ops.erase(Ops.begin()+Idx);
781 Ops.erase(Ops.begin()+AddOp-1);
783 Ops.push_back(OuterMul);
784 return getAddExpr(Ops);
787 // Check this multiply against other multiplies being added together.
788 for (unsigned OtherMulIdx = Idx+1;
789 OtherMulIdx < Ops.size() && isa<SCEVMulExpr>(Ops[OtherMulIdx]);
791 SCEVMulExpr *OtherMul = cast<SCEVMulExpr>(Ops[OtherMulIdx]);
792 // If MulOp occurs in OtherMul, we can fold the two multiplies
794 for (unsigned OMulOp = 0, e = OtherMul->getNumOperands();
795 OMulOp != e; ++OMulOp)
796 if (OtherMul->getOperand(OMulOp) == MulOpSCEV) {
797 // Fold X + (A*B*C) + (A*D*E) --> X + (A*(B*C+D*E))
798 SCEVHandle InnerMul1 = Mul->getOperand(MulOp == 0);
799 if (Mul->getNumOperands() != 2) {
800 std::vector<SCEVHandle> MulOps(Mul->op_begin(), Mul->op_end());
801 MulOps.erase(MulOps.begin()+MulOp);
802 InnerMul1 = getMulExpr(MulOps);
804 SCEVHandle InnerMul2 = OtherMul->getOperand(OMulOp == 0);
805 if (OtherMul->getNumOperands() != 2) {
806 std::vector<SCEVHandle> MulOps(OtherMul->op_begin(),
808 MulOps.erase(MulOps.begin()+OMulOp);
809 InnerMul2 = getMulExpr(MulOps);
811 SCEVHandle InnerMulSum = getAddExpr(InnerMul1,InnerMul2);
812 SCEVHandle OuterMul = getMulExpr(MulOpSCEV, InnerMulSum);
813 if (Ops.size() == 2) return OuterMul;
814 Ops.erase(Ops.begin()+Idx);
815 Ops.erase(Ops.begin()+OtherMulIdx-1);
816 Ops.push_back(OuterMul);
817 return getAddExpr(Ops);
823 // If there are any add recurrences in the operands list, see if any other
824 // added values are loop invariant. If so, we can fold them into the
826 while (Idx < Ops.size() && Ops[Idx]->getSCEVType() < scAddRecExpr)
829 // Scan over all recurrences, trying to fold loop invariants into them.
830 for (; Idx < Ops.size() && isa<SCEVAddRecExpr>(Ops[Idx]); ++Idx) {
831 // Scan all of the other operands to this add and add them to the vector if
832 // they are loop invariant w.r.t. the recurrence.
833 std::vector<SCEVHandle> LIOps;
834 SCEVAddRecExpr *AddRec = cast<SCEVAddRecExpr>(Ops[Idx]);
835 for (unsigned i = 0, e = Ops.size(); i != e; ++i)
836 if (Ops[i]->isLoopInvariant(AddRec->getLoop())) {
837 LIOps.push_back(Ops[i]);
838 Ops.erase(Ops.begin()+i);
842 // If we found some loop invariants, fold them into the recurrence.
843 if (!LIOps.empty()) {
844 // NLI + LI + { Start,+,Step} --> NLI + { LI+Start,+,Step }
845 LIOps.push_back(AddRec->getStart());
847 std::vector<SCEVHandle> AddRecOps(AddRec->op_begin(), AddRec->op_end());
848 AddRecOps[0] = getAddExpr(LIOps);
850 SCEVHandle NewRec = getAddRecExpr(AddRecOps, AddRec->getLoop());
851 // If all of the other operands were loop invariant, we are done.
852 if (Ops.size() == 1) return NewRec;
854 // Otherwise, add the folded AddRec by the non-liv parts.
855 for (unsigned i = 0;; ++i)
856 if (Ops[i] == AddRec) {
860 return getAddExpr(Ops);
863 // Okay, if there weren't any loop invariants to be folded, check to see if
864 // there are multiple AddRec's with the same loop induction variable being
865 // added together. If so, we can fold them.
866 for (unsigned OtherIdx = Idx+1;
867 OtherIdx < Ops.size() && isa<SCEVAddRecExpr>(Ops[OtherIdx]);++OtherIdx)
868 if (OtherIdx != Idx) {
869 SCEVAddRecExpr *OtherAddRec = cast<SCEVAddRecExpr>(Ops[OtherIdx]);
870 if (AddRec->getLoop() == OtherAddRec->getLoop()) {
871 // Other + {A,+,B} + {C,+,D} --> Other + {A+C,+,B+D}
872 std::vector<SCEVHandle> NewOps(AddRec->op_begin(), AddRec->op_end());
873 for (unsigned i = 0, e = OtherAddRec->getNumOperands(); i != e; ++i) {
874 if (i >= NewOps.size()) {
875 NewOps.insert(NewOps.end(), OtherAddRec->op_begin()+i,
876 OtherAddRec->op_end());
879 NewOps[i] = getAddExpr(NewOps[i], OtherAddRec->getOperand(i));
881 SCEVHandle NewAddRec = getAddRecExpr(NewOps, AddRec->getLoop());
883 if (Ops.size() == 2) return NewAddRec;
885 Ops.erase(Ops.begin()+Idx);
886 Ops.erase(Ops.begin()+OtherIdx-1);
887 Ops.push_back(NewAddRec);
888 return getAddExpr(Ops);
892 // Otherwise couldn't fold anything into this recurrence. Move onto the
896 // Okay, it looks like we really DO need an add expr. Check to see if we
897 // already have one, otherwise create a new one.
898 std::vector<SCEV*> SCEVOps(Ops.begin(), Ops.end());
899 SCEVCommutativeExpr *&Result = (*SCEVCommExprs)[std::make_pair(scAddExpr,
901 if (Result == 0) Result = new SCEVAddExpr(Ops);
906 SCEVHandle ScalarEvolution::getMulExpr(std::vector<SCEVHandle> &Ops) {
907 assert(!Ops.empty() && "Cannot get empty mul!");
909 // Sort by complexity, this groups all similar expression types together.
910 GroupByComplexity(Ops);
912 // If there are any constants, fold them together.
914 if (SCEVConstant *LHSC = dyn_cast<SCEVConstant>(Ops[0])) {
916 // C1*(C2+V) -> C1*C2 + C1*V
918 if (SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(Ops[1]))
919 if (Add->getNumOperands() == 2 &&
920 isa<SCEVConstant>(Add->getOperand(0)))
921 return getAddExpr(getMulExpr(LHSC, Add->getOperand(0)),
922 getMulExpr(LHSC, Add->getOperand(1)));
926 while (SCEVConstant *RHSC = dyn_cast<SCEVConstant>(Ops[Idx])) {
927 // We found two constants, fold them together!
928 ConstantInt *Fold = ConstantInt::get(LHSC->getValue()->getValue() *
929 RHSC->getValue()->getValue());
930 Ops[0] = getConstant(Fold);
931 Ops.erase(Ops.begin()+1); // Erase the folded element
932 if (Ops.size() == 1) return Ops[0];
933 LHSC = cast<SCEVConstant>(Ops[0]);
936 // If we are left with a constant one being multiplied, strip it off.
937 if (cast<SCEVConstant>(Ops[0])->getValue()->equalsInt(1)) {
938 Ops.erase(Ops.begin());
940 } else if (cast<SCEVConstant>(Ops[0])->getValue()->isZero()) {
941 // If we have a multiply of zero, it will always be zero.
946 // Skip over the add expression until we get to a multiply.
947 while (Idx < Ops.size() && Ops[Idx]->getSCEVType() < scMulExpr)
953 // If there are mul operands inline them all into this expression.
954 if (Idx < Ops.size()) {
955 bool DeletedMul = false;
956 while (SCEVMulExpr *Mul = dyn_cast<SCEVMulExpr>(Ops[Idx])) {
957 // If we have an mul, expand the mul operands onto the end of the operands
959 Ops.insert(Ops.end(), Mul->op_begin(), Mul->op_end());
960 Ops.erase(Ops.begin()+Idx);
964 // If we deleted at least one mul, we added operands to the end of the list,
965 // and they are not necessarily sorted. Recurse to resort and resimplify
966 // any operands we just aquired.
968 return getMulExpr(Ops);
971 // If there are any add recurrences in the operands list, see if any other
972 // added values are loop invariant. If so, we can fold them into the
974 while (Idx < Ops.size() && Ops[Idx]->getSCEVType() < scAddRecExpr)
977 // Scan over all recurrences, trying to fold loop invariants into them.
978 for (; Idx < Ops.size() && isa<SCEVAddRecExpr>(Ops[Idx]); ++Idx) {
979 // Scan all of the other operands to this mul and add them to the vector if
980 // they are loop invariant w.r.t. the recurrence.
981 std::vector<SCEVHandle> LIOps;
982 SCEVAddRecExpr *AddRec = cast<SCEVAddRecExpr>(Ops[Idx]);
983 for (unsigned i = 0, e = Ops.size(); i != e; ++i)
984 if (Ops[i]->isLoopInvariant(AddRec->getLoop())) {
985 LIOps.push_back(Ops[i]);
986 Ops.erase(Ops.begin()+i);
990 // If we found some loop invariants, fold them into the recurrence.
991 if (!LIOps.empty()) {
992 // NLI * LI * { Start,+,Step} --> NLI * { LI*Start,+,LI*Step }
993 std::vector<SCEVHandle> NewOps;
994 NewOps.reserve(AddRec->getNumOperands());
995 if (LIOps.size() == 1) {
996 SCEV *Scale = LIOps[0];
997 for (unsigned i = 0, e = AddRec->getNumOperands(); i != e; ++i)
998 NewOps.push_back(getMulExpr(Scale, AddRec->getOperand(i)));
1000 for (unsigned i = 0, e = AddRec->getNumOperands(); i != e; ++i) {
1001 std::vector<SCEVHandle> MulOps(LIOps);
1002 MulOps.push_back(AddRec->getOperand(i));
1003 NewOps.push_back(getMulExpr(MulOps));
1007 SCEVHandle NewRec = getAddRecExpr(NewOps, AddRec->getLoop());
1009 // If all of the other operands were loop invariant, we are done.
1010 if (Ops.size() == 1) return NewRec;
1012 // Otherwise, multiply the folded AddRec by the non-liv parts.
1013 for (unsigned i = 0;; ++i)
1014 if (Ops[i] == AddRec) {
1018 return getMulExpr(Ops);
1021 // Okay, if there weren't any loop invariants to be folded, check to see if
1022 // there are multiple AddRec's with the same loop induction variable being
1023 // multiplied together. If so, we can fold them.
1024 for (unsigned OtherIdx = Idx+1;
1025 OtherIdx < Ops.size() && isa<SCEVAddRecExpr>(Ops[OtherIdx]);++OtherIdx)
1026 if (OtherIdx != Idx) {
1027 SCEVAddRecExpr *OtherAddRec = cast<SCEVAddRecExpr>(Ops[OtherIdx]);
1028 if (AddRec->getLoop() == OtherAddRec->getLoop()) {
1029 // F * G --> {A,+,B} * {C,+,D} --> {A*C,+,F*D + G*B + B*D}
1030 SCEVAddRecExpr *F = AddRec, *G = OtherAddRec;
1031 SCEVHandle NewStart = getMulExpr(F->getStart(),
1033 SCEVHandle B = F->getStepRecurrence(*this);
1034 SCEVHandle D = G->getStepRecurrence(*this);
1035 SCEVHandle NewStep = getAddExpr(getMulExpr(F, D),
1038 SCEVHandle NewAddRec = getAddRecExpr(NewStart, NewStep,
1040 if (Ops.size() == 2) return NewAddRec;
1042 Ops.erase(Ops.begin()+Idx);
1043 Ops.erase(Ops.begin()+OtherIdx-1);
1044 Ops.push_back(NewAddRec);
1045 return getMulExpr(Ops);
1049 // Otherwise couldn't fold anything into this recurrence. Move onto the
1053 // Okay, it looks like we really DO need an mul expr. Check to see if we
1054 // already have one, otherwise create a new one.
1055 std::vector<SCEV*> SCEVOps(Ops.begin(), Ops.end());
1056 SCEVCommutativeExpr *&Result = (*SCEVCommExprs)[std::make_pair(scMulExpr,
1059 Result = new SCEVMulExpr(Ops);
1063 SCEVHandle ScalarEvolution::getUDivExpr(const SCEVHandle &LHS, const SCEVHandle &RHS) {
1064 if (SCEVConstant *RHSC = dyn_cast<SCEVConstant>(RHS)) {
1065 if (RHSC->getValue()->equalsInt(1))
1066 return LHS; // X udiv 1 --> x
1068 if (SCEVConstant *LHSC = dyn_cast<SCEVConstant>(LHS)) {
1069 Constant *LHSCV = LHSC->getValue();
1070 Constant *RHSCV = RHSC->getValue();
1071 return getUnknown(ConstantExpr::getUDiv(LHSCV, RHSCV));
1075 // FIXME: implement folding of (X*4)/4 when we know X*4 doesn't overflow.
1077 SCEVUDivExpr *&Result = (*SCEVUDivs)[std::make_pair(LHS, RHS)];
1078 if (Result == 0) Result = new SCEVUDivExpr(LHS, RHS);
1083 /// SCEVAddRecExpr::get - Get a add recurrence expression for the
1084 /// specified loop. Simplify the expression as much as possible.
1085 SCEVHandle ScalarEvolution::getAddRecExpr(const SCEVHandle &Start,
1086 const SCEVHandle &Step, const Loop *L) {
1087 std::vector<SCEVHandle> Operands;
1088 Operands.push_back(Start);
1089 if (SCEVAddRecExpr *StepChrec = dyn_cast<SCEVAddRecExpr>(Step))
1090 if (StepChrec->getLoop() == L) {
1091 Operands.insert(Operands.end(), StepChrec->op_begin(),
1092 StepChrec->op_end());
1093 return getAddRecExpr(Operands, L);
1096 Operands.push_back(Step);
1097 return getAddRecExpr(Operands, L);
1100 /// SCEVAddRecExpr::get - Get a add recurrence expression for the
1101 /// specified loop. Simplify the expression as much as possible.
1102 SCEVHandle ScalarEvolution::getAddRecExpr(std::vector<SCEVHandle> &Operands,
1104 if (Operands.size() == 1) return Operands[0];
1106 if (Operands.back()->isZero()) {
1107 Operands.pop_back();
1108 return getAddRecExpr(Operands, L); // { X,+,0 } --> X
1111 SCEVAddRecExpr *&Result =
1112 (*SCEVAddRecExprs)[std::make_pair(L, std::vector<SCEV*>(Operands.begin(),
1114 if (Result == 0) Result = new SCEVAddRecExpr(Operands, L);
1118 SCEVHandle ScalarEvolution::getSMaxExpr(const SCEVHandle &LHS,
1119 const SCEVHandle &RHS) {
1120 std::vector<SCEVHandle> Ops;
1123 return getSMaxExpr(Ops);
1126 SCEVHandle ScalarEvolution::getSMaxExpr(std::vector<SCEVHandle> Ops) {
1127 assert(!Ops.empty() && "Cannot get empty smax!");
1128 if (Ops.size() == 1) return Ops[0];
1130 // Sort by complexity, this groups all similar expression types together.
1131 GroupByComplexity(Ops);
1133 // If there are any constants, fold them together.
1135 if (SCEVConstant *LHSC = dyn_cast<SCEVConstant>(Ops[0])) {
1137 assert(Idx < Ops.size());
1138 while (SCEVConstant *RHSC = dyn_cast<SCEVConstant>(Ops[Idx])) {
1139 // We found two constants, fold them together!
1140 ConstantInt *Fold = ConstantInt::get(
1141 APIntOps::smax(LHSC->getValue()->getValue(),
1142 RHSC->getValue()->getValue()));
1143 Ops[0] = getConstant(Fold);
1144 Ops.erase(Ops.begin()+1); // Erase the folded element
1145 if (Ops.size() == 1) return Ops[0];
1146 LHSC = cast<SCEVConstant>(Ops[0]);
1149 // If we are left with a constant -inf, strip it off.
1150 if (cast<SCEVConstant>(Ops[0])->getValue()->isMinValue(true)) {
1151 Ops.erase(Ops.begin());
1156 if (Ops.size() == 1) return Ops[0];
1158 // Find the first SMax
1159 while (Idx < Ops.size() && Ops[Idx]->getSCEVType() < scSMaxExpr)
1162 // Check to see if one of the operands is an SMax. If so, expand its operands
1163 // onto our operand list, and recurse to simplify.
1164 if (Idx < Ops.size()) {
1165 bool DeletedSMax = false;
1166 while (SCEVSMaxExpr *SMax = dyn_cast<SCEVSMaxExpr>(Ops[Idx])) {
1167 Ops.insert(Ops.end(), SMax->op_begin(), SMax->op_end());
1168 Ops.erase(Ops.begin()+Idx);
1173 return getSMaxExpr(Ops);
1176 // Okay, check to see if the same value occurs in the operand list twice. If
1177 // so, delete one. Since we sorted the list, these values are required to
1179 for (unsigned i = 0, e = Ops.size()-1; i != e; ++i)
1180 if (Ops[i] == Ops[i+1]) { // X smax Y smax Y --> X smax Y
1181 Ops.erase(Ops.begin()+i, Ops.begin()+i+1);
1185 if (Ops.size() == 1) return Ops[0];
1187 assert(!Ops.empty() && "Reduced smax down to nothing!");
1189 // Okay, it looks like we really DO need an smax expr. Check to see if we
1190 // already have one, otherwise create a new one.
1191 std::vector<SCEV*> SCEVOps(Ops.begin(), Ops.end());
1192 SCEVCommutativeExpr *&Result = (*SCEVCommExprs)[std::make_pair(scSMaxExpr,
1194 if (Result == 0) Result = new SCEVSMaxExpr(Ops);
1198 SCEVHandle ScalarEvolution::getUMaxExpr(const SCEVHandle &LHS,
1199 const SCEVHandle &RHS) {
1200 std::vector<SCEVHandle> Ops;
1203 return getUMaxExpr(Ops);
1206 SCEVHandle ScalarEvolution::getUMaxExpr(std::vector<SCEVHandle> Ops) {
1207 assert(!Ops.empty() && "Cannot get empty umax!");
1208 if (Ops.size() == 1) return Ops[0];
1210 // Sort by complexity, this groups all similar expression types together.
1211 GroupByComplexity(Ops);
1213 // If there are any constants, fold them together.
1215 if (SCEVConstant *LHSC = dyn_cast<SCEVConstant>(Ops[0])) {
1217 assert(Idx < Ops.size());
1218 while (SCEVConstant *RHSC = dyn_cast<SCEVConstant>(Ops[Idx])) {
1219 // We found two constants, fold them together!
1220 ConstantInt *Fold = ConstantInt::get(
1221 APIntOps::umax(LHSC->getValue()->getValue(),
1222 RHSC->getValue()->getValue()));
1223 Ops[0] = getConstant(Fold);
1224 Ops.erase(Ops.begin()+1); // Erase the folded element
1225 if (Ops.size() == 1) return Ops[0];
1226 LHSC = cast<SCEVConstant>(Ops[0]);
1229 // If we are left with a constant zero, strip it off.
1230 if (cast<SCEVConstant>(Ops[0])->getValue()->isMinValue(false)) {
1231 Ops.erase(Ops.begin());
1236 if (Ops.size() == 1) return Ops[0];
1238 // Find the first UMax
1239 while (Idx < Ops.size() && Ops[Idx]->getSCEVType() < scUMaxExpr)
1242 // Check to see if one of the operands is a UMax. If so, expand its operands
1243 // onto our operand list, and recurse to simplify.
1244 if (Idx < Ops.size()) {
1245 bool DeletedUMax = false;
1246 while (SCEVUMaxExpr *UMax = dyn_cast<SCEVUMaxExpr>(Ops[Idx])) {
1247 Ops.insert(Ops.end(), UMax->op_begin(), UMax->op_end());
1248 Ops.erase(Ops.begin()+Idx);
1253 return getUMaxExpr(Ops);
1256 // Okay, check to see if the same value occurs in the operand list twice. If
1257 // so, delete one. Since we sorted the list, these values are required to
1259 for (unsigned i = 0, e = Ops.size()-1; i != e; ++i)
1260 if (Ops[i] == Ops[i+1]) { // X umax Y umax Y --> X umax Y
1261 Ops.erase(Ops.begin()+i, Ops.begin()+i+1);
1265 if (Ops.size() == 1) return Ops[0];
1267 assert(!Ops.empty() && "Reduced umax down to nothing!");
1269 // Okay, it looks like we really DO need a umax expr. Check to see if we
1270 // already have one, otherwise create a new one.
1271 std::vector<SCEV*> SCEVOps(Ops.begin(), Ops.end());
1272 SCEVCommutativeExpr *&Result = (*SCEVCommExprs)[std::make_pair(scUMaxExpr,
1274 if (Result == 0) Result = new SCEVUMaxExpr(Ops);
1278 SCEVHandle ScalarEvolution::getUnknown(Value *V) {
1279 if (ConstantInt *CI = dyn_cast<ConstantInt>(V))
1280 return getConstant(CI);
1281 SCEVUnknown *&Result = (*SCEVUnknowns)[V];
1282 if (Result == 0) Result = new SCEVUnknown(V);
1287 //===----------------------------------------------------------------------===//
1288 // ScalarEvolutionsImpl Definition and Implementation
1289 //===----------------------------------------------------------------------===//
1291 /// ScalarEvolutionsImpl - This class implements the main driver for the scalar
1295 struct VISIBILITY_HIDDEN ScalarEvolutionsImpl {
1296 /// SE - A reference to the public ScalarEvolution object.
1297 ScalarEvolution &SE;
1299 /// F - The function we are analyzing.
1303 /// LI - The loop information for the function we are currently analyzing.
1307 /// UnknownValue - This SCEV is used to represent unknown trip counts and
1309 SCEVHandle UnknownValue;
1311 /// Scalars - This is a cache of the scalars we have analyzed so far.
1313 std::map<Value*, SCEVHandle> Scalars;
1315 /// IterationCounts - Cache the iteration count of the loops for this
1316 /// function as they are computed.
1317 std::map<const Loop*, SCEVHandle> IterationCounts;
1319 /// ConstantEvolutionLoopExitValue - This map contains entries for all of
1320 /// the PHI instructions that we attempt to compute constant evolutions for.
1321 /// This allows us to avoid potentially expensive recomputation of these
1322 /// properties. An instruction maps to null if we are unable to compute its
1324 std::map<PHINode*, Constant*> ConstantEvolutionLoopExitValue;
1327 ScalarEvolutionsImpl(ScalarEvolution &se, Function &f, LoopInfo &li)
1328 : SE(se), F(f), LI(li), UnknownValue(new SCEVCouldNotCompute()) {}
1330 /// getSCEV - Return an existing SCEV if it exists, otherwise analyze the
1331 /// expression and create a new one.
1332 SCEVHandle getSCEV(Value *V);
1334 /// hasSCEV - Return true if the SCEV for this value has already been
1336 bool hasSCEV(Value *V) const {
1337 return Scalars.count(V);
1340 /// setSCEV - Insert the specified SCEV into the map of current SCEVs for
1341 /// the specified value.
1342 void setSCEV(Value *V, const SCEVHandle &H) {
1343 bool isNew = Scalars.insert(std::make_pair(V, H)).second;
1344 assert(isNew && "This entry already existed!");
1348 /// getSCEVAtScope - Compute the value of the specified expression within
1349 /// the indicated loop (which may be null to indicate in no loop). If the
1350 /// expression cannot be evaluated, return UnknownValue itself.
1351 SCEVHandle getSCEVAtScope(SCEV *V, const Loop *L);
1354 /// hasLoopInvariantIterationCount - Return true if the specified loop has
1355 /// an analyzable loop-invariant iteration count.
1356 bool hasLoopInvariantIterationCount(const Loop *L);
1358 /// getIterationCount - If the specified loop has a predictable iteration
1359 /// count, return it. Note that it is not valid to call this method on a
1360 /// loop without a loop-invariant iteration count.
1361 SCEVHandle getIterationCount(const Loop *L);
1363 /// deleteValueFromRecords - This method should be called by the
1364 /// client before it removes a value from the program, to make sure
1365 /// that no dangling references are left around.
1366 void deleteValueFromRecords(Value *V);
1369 /// createSCEV - We know that there is no SCEV for the specified value.
1370 /// Analyze the expression.
1371 SCEVHandle createSCEV(Value *V);
1373 /// createNodeForPHI - Provide the special handling we need to analyze PHI
1375 SCEVHandle createNodeForPHI(PHINode *PN);
1377 /// ReplaceSymbolicValueWithConcrete - This looks up the computed SCEV value
1378 /// for the specified instruction and replaces any references to the
1379 /// symbolic value SymName with the specified value. This is used during
1381 void ReplaceSymbolicValueWithConcrete(Instruction *I,
1382 const SCEVHandle &SymName,
1383 const SCEVHandle &NewVal);
1385 /// ComputeIterationCount - Compute the number of times the specified loop
1387 SCEVHandle ComputeIterationCount(const Loop *L);
1389 /// ComputeLoadConstantCompareIterationCount - Given an exit condition of
1390 /// 'icmp op load X, cst', try to see if we can compute the trip count.
1391 SCEVHandle ComputeLoadConstantCompareIterationCount(LoadInst *LI,
1394 ICmpInst::Predicate p);
1396 /// ComputeIterationCountExhaustively - If the trip is known to execute a
1397 /// constant number of times (the condition evolves only from constants),
1398 /// try to evaluate a few iterations of the loop until we get the exit
1399 /// condition gets a value of ExitWhen (true or false). If we cannot
1400 /// evaluate the trip count of the loop, return UnknownValue.
1401 SCEVHandle ComputeIterationCountExhaustively(const Loop *L, Value *Cond,
1404 /// HowFarToZero - Return the number of times a backedge comparing the
1405 /// specified value to zero will execute. If not computable, return
1407 SCEVHandle HowFarToZero(SCEV *V, const Loop *L);
1409 /// HowFarToNonZero - Return the number of times a backedge checking the
1410 /// specified value for nonzero will execute. If not computable, return
1412 SCEVHandle HowFarToNonZero(SCEV *V, const Loop *L);
1414 /// HowManyLessThans - Return the number of times a backedge containing the
1415 /// specified less-than comparison will execute. If not computable, return
1416 /// UnknownValue. isSigned specifies whether the less-than is signed.
1417 SCEVHandle HowManyLessThans(SCEV *LHS, SCEV *RHS, const Loop *L,
1420 /// executesAtLeastOnce - Test whether entry to the loop is protected by
1421 /// a conditional between LHS and RHS.
1422 bool executesAtLeastOnce(const Loop *L, bool isSigned, SCEV *LHS, SCEV *RHS);
1424 /// getConstantEvolutionLoopExitValue - If we know that the specified Phi is
1425 /// in the header of its containing loop, we know the loop executes a
1426 /// constant number of times, and the PHI node is just a recurrence
1427 /// involving constants, fold it.
1428 Constant *getConstantEvolutionLoopExitValue(PHINode *PN, const APInt& Its,
1433 //===----------------------------------------------------------------------===//
1434 // Basic SCEV Analysis and PHI Idiom Recognition Code
1437 /// deleteValueFromRecords - This method should be called by the
1438 /// client before it removes an instruction from the program, to make sure
1439 /// that no dangling references are left around.
1440 void ScalarEvolutionsImpl::deleteValueFromRecords(Value *V) {
1441 SmallVector<Value *, 16> Worklist;
1443 if (Scalars.erase(V)) {
1444 if (PHINode *PN = dyn_cast<PHINode>(V))
1445 ConstantEvolutionLoopExitValue.erase(PN);
1446 Worklist.push_back(V);
1449 while (!Worklist.empty()) {
1450 Value *VV = Worklist.back();
1451 Worklist.pop_back();
1453 for (Instruction::use_iterator UI = VV->use_begin(), UE = VV->use_end();
1455 Instruction *Inst = cast<Instruction>(*UI);
1456 if (Scalars.erase(Inst)) {
1457 if (PHINode *PN = dyn_cast<PHINode>(VV))
1458 ConstantEvolutionLoopExitValue.erase(PN);
1459 Worklist.push_back(Inst);
1466 /// getSCEV - Return an existing SCEV if it exists, otherwise analyze the
1467 /// expression and create a new one.
1468 SCEVHandle ScalarEvolutionsImpl::getSCEV(Value *V) {
1469 assert(V->getType() != Type::VoidTy && "Can't analyze void expressions!");
1471 std::map<Value*, SCEVHandle>::iterator I = Scalars.find(V);
1472 if (I != Scalars.end()) return I->second;
1473 SCEVHandle S = createSCEV(V);
1474 Scalars.insert(std::make_pair(V, S));
1478 /// ReplaceSymbolicValueWithConcrete - This looks up the computed SCEV value for
1479 /// the specified instruction and replaces any references to the symbolic value
1480 /// SymName with the specified value. This is used during PHI resolution.
1481 void ScalarEvolutionsImpl::
1482 ReplaceSymbolicValueWithConcrete(Instruction *I, const SCEVHandle &SymName,
1483 const SCEVHandle &NewVal) {
1484 std::map<Value*, SCEVHandle>::iterator SI = Scalars.find(I);
1485 if (SI == Scalars.end()) return;
1488 SI->second->replaceSymbolicValuesWithConcrete(SymName, NewVal, SE);
1489 if (NV == SI->second) return; // No change.
1491 SI->second = NV; // Update the scalars map!
1493 // Any instruction values that use this instruction might also need to be
1495 for (Value::use_iterator UI = I->use_begin(), E = I->use_end();
1497 ReplaceSymbolicValueWithConcrete(cast<Instruction>(*UI), SymName, NewVal);
1500 /// createNodeForPHI - PHI nodes have two cases. Either the PHI node exists in
1501 /// a loop header, making it a potential recurrence, or it doesn't.
1503 SCEVHandle ScalarEvolutionsImpl::createNodeForPHI(PHINode *PN) {
1504 if (PN->getNumIncomingValues() == 2) // The loops have been canonicalized.
1505 if (const Loop *L = LI.getLoopFor(PN->getParent()))
1506 if (L->getHeader() == PN->getParent()) {
1507 // If it lives in the loop header, it has two incoming values, one
1508 // from outside the loop, and one from inside.
1509 unsigned IncomingEdge = L->contains(PN->getIncomingBlock(0));
1510 unsigned BackEdge = IncomingEdge^1;
1512 // While we are analyzing this PHI node, handle its value symbolically.
1513 SCEVHandle SymbolicName = SE.getUnknown(PN);
1514 assert(Scalars.find(PN) == Scalars.end() &&
1515 "PHI node already processed?");
1516 Scalars.insert(std::make_pair(PN, SymbolicName));
1518 // Using this symbolic name for the PHI, analyze the value coming around
1520 SCEVHandle BEValue = getSCEV(PN->getIncomingValue(BackEdge));
1522 // NOTE: If BEValue is loop invariant, we know that the PHI node just
1523 // has a special value for the first iteration of the loop.
1525 // If the value coming around the backedge is an add with the symbolic
1526 // value we just inserted, then we found a simple induction variable!
1527 if (SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(BEValue)) {
1528 // If there is a single occurrence of the symbolic value, replace it
1529 // with a recurrence.
1530 unsigned FoundIndex = Add->getNumOperands();
1531 for (unsigned i = 0, e = Add->getNumOperands(); i != e; ++i)
1532 if (Add->getOperand(i) == SymbolicName)
1533 if (FoundIndex == e) {
1538 if (FoundIndex != Add->getNumOperands()) {
1539 // Create an add with everything but the specified operand.
1540 std::vector<SCEVHandle> Ops;
1541 for (unsigned i = 0, e = Add->getNumOperands(); i != e; ++i)
1542 if (i != FoundIndex)
1543 Ops.push_back(Add->getOperand(i));
1544 SCEVHandle Accum = SE.getAddExpr(Ops);
1546 // This is not a valid addrec if the step amount is varying each
1547 // loop iteration, but is not itself an addrec in this loop.
1548 if (Accum->isLoopInvariant(L) ||
1549 (isa<SCEVAddRecExpr>(Accum) &&
1550 cast<SCEVAddRecExpr>(Accum)->getLoop() == L)) {
1551 SCEVHandle StartVal = getSCEV(PN->getIncomingValue(IncomingEdge));
1552 SCEVHandle PHISCEV = SE.getAddRecExpr(StartVal, Accum, L);
1554 // Okay, for the entire analysis of this edge we assumed the PHI
1555 // to be symbolic. We now need to go back and update all of the
1556 // entries for the scalars that use the PHI (except for the PHI
1557 // itself) to use the new analyzed value instead of the "symbolic"
1559 ReplaceSymbolicValueWithConcrete(PN, SymbolicName, PHISCEV);
1563 } else if (SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(BEValue)) {
1564 // Otherwise, this could be a loop like this:
1565 // i = 0; for (j = 1; ..; ++j) { .... i = j; }
1566 // In this case, j = {1,+,1} and BEValue is j.
1567 // Because the other in-value of i (0) fits the evolution of BEValue
1568 // i really is an addrec evolution.
1569 if (AddRec->getLoop() == L && AddRec->isAffine()) {
1570 SCEVHandle StartVal = getSCEV(PN->getIncomingValue(IncomingEdge));
1572 // If StartVal = j.start - j.stride, we can use StartVal as the
1573 // initial step of the addrec evolution.
1574 if (StartVal == SE.getMinusSCEV(AddRec->getOperand(0),
1575 AddRec->getOperand(1))) {
1576 SCEVHandle PHISCEV =
1577 SE.getAddRecExpr(StartVal, AddRec->getOperand(1), L);
1579 // Okay, for the entire analysis of this edge we assumed the PHI
1580 // to be symbolic. We now need to go back and update all of the
1581 // entries for the scalars that use the PHI (except for the PHI
1582 // itself) to use the new analyzed value instead of the "symbolic"
1584 ReplaceSymbolicValueWithConcrete(PN, SymbolicName, PHISCEV);
1590 return SymbolicName;
1593 // If it's not a loop phi, we can't handle it yet.
1594 return SE.getUnknown(PN);
1597 /// GetMinTrailingZeros - Determine the minimum number of zero bits that S is
1598 /// guaranteed to end in (at every loop iteration). It is, at the same time,
1599 /// the minimum number of times S is divisible by 2. For example, given {4,+,8}
1600 /// it returns 2. If S is guaranteed to be 0, it returns the bitwidth of S.
1601 static uint32_t GetMinTrailingZeros(SCEVHandle S) {
1602 if (SCEVConstant *C = dyn_cast<SCEVConstant>(S))
1603 return C->getValue()->getValue().countTrailingZeros();
1605 if (SCEVTruncateExpr *T = dyn_cast<SCEVTruncateExpr>(S))
1606 return std::min(GetMinTrailingZeros(T->getOperand()), T->getBitWidth());
1608 if (SCEVZeroExtendExpr *E = dyn_cast<SCEVZeroExtendExpr>(S)) {
1609 uint32_t OpRes = GetMinTrailingZeros(E->getOperand());
1610 return OpRes == E->getOperand()->getBitWidth() ? E->getBitWidth() : OpRes;
1613 if (SCEVSignExtendExpr *E = dyn_cast<SCEVSignExtendExpr>(S)) {
1614 uint32_t OpRes = GetMinTrailingZeros(E->getOperand());
1615 return OpRes == E->getOperand()->getBitWidth() ? E->getBitWidth() : OpRes;
1618 if (SCEVAddExpr *A = dyn_cast<SCEVAddExpr>(S)) {
1619 // The result is the min of all operands results.
1620 uint32_t MinOpRes = GetMinTrailingZeros(A->getOperand(0));
1621 for (unsigned i = 1, e = A->getNumOperands(); MinOpRes && i != e; ++i)
1622 MinOpRes = std::min(MinOpRes, GetMinTrailingZeros(A->getOperand(i)));
1626 if (SCEVMulExpr *M = dyn_cast<SCEVMulExpr>(S)) {
1627 // The result is the sum of all operands results.
1628 uint32_t SumOpRes = GetMinTrailingZeros(M->getOperand(0));
1629 uint32_t BitWidth = M->getBitWidth();
1630 for (unsigned i = 1, e = M->getNumOperands();
1631 SumOpRes != BitWidth && i != e; ++i)
1632 SumOpRes = std::min(SumOpRes + GetMinTrailingZeros(M->getOperand(i)),
1637 if (SCEVAddRecExpr *A = dyn_cast<SCEVAddRecExpr>(S)) {
1638 // The result is the min of all operands results.
1639 uint32_t MinOpRes = GetMinTrailingZeros(A->getOperand(0));
1640 for (unsigned i = 1, e = A->getNumOperands(); MinOpRes && i != e; ++i)
1641 MinOpRes = std::min(MinOpRes, GetMinTrailingZeros(A->getOperand(i)));
1645 if (SCEVSMaxExpr *M = dyn_cast<SCEVSMaxExpr>(S)) {
1646 // The result is the min of all operands results.
1647 uint32_t MinOpRes = GetMinTrailingZeros(M->getOperand(0));
1648 for (unsigned i = 1, e = M->getNumOperands(); MinOpRes && i != e; ++i)
1649 MinOpRes = std::min(MinOpRes, GetMinTrailingZeros(M->getOperand(i)));
1653 if (SCEVUMaxExpr *M = dyn_cast<SCEVUMaxExpr>(S)) {
1654 // The result is the min of all operands results.
1655 uint32_t MinOpRes = GetMinTrailingZeros(M->getOperand(0));
1656 for (unsigned i = 1, e = M->getNumOperands(); MinOpRes && i != e; ++i)
1657 MinOpRes = std::min(MinOpRes, GetMinTrailingZeros(M->getOperand(i)));
1661 // SCEVUDivExpr, SCEVUnknown
1665 /// createSCEV - We know that there is no SCEV for the specified value.
1666 /// Analyze the expression.
1668 SCEVHandle ScalarEvolutionsImpl::createSCEV(Value *V) {
1669 if (!isa<IntegerType>(V->getType()))
1670 return SE.getUnknown(V);
1672 unsigned Opcode = Instruction::UserOp1;
1673 if (Instruction *I = dyn_cast<Instruction>(V))
1674 Opcode = I->getOpcode();
1675 else if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V))
1676 Opcode = CE->getOpcode();
1678 return SE.getUnknown(V);
1680 User *U = cast<User>(V);
1682 case Instruction::Add:
1683 return SE.getAddExpr(getSCEV(U->getOperand(0)),
1684 getSCEV(U->getOperand(1)));
1685 case Instruction::Mul:
1686 return SE.getMulExpr(getSCEV(U->getOperand(0)),
1687 getSCEV(U->getOperand(1)));
1688 case Instruction::UDiv:
1689 return SE.getUDivExpr(getSCEV(U->getOperand(0)),
1690 getSCEV(U->getOperand(1)));
1691 case Instruction::Sub:
1692 return SE.getMinusSCEV(getSCEV(U->getOperand(0)),
1693 getSCEV(U->getOperand(1)));
1694 case Instruction::Or:
1695 // If the RHS of the Or is a constant, we may have something like:
1696 // X*4+1 which got turned into X*4|1. Handle this as an Add so loop
1697 // optimizations will transparently handle this case.
1699 // In order for this transformation to be safe, the LHS must be of the
1700 // form X*(2^n) and the Or constant must be less than 2^n.
1701 if (ConstantInt *CI = dyn_cast<ConstantInt>(U->getOperand(1))) {
1702 SCEVHandle LHS = getSCEV(U->getOperand(0));
1703 const APInt &CIVal = CI->getValue();
1704 if (GetMinTrailingZeros(LHS) >=
1705 (CIVal.getBitWidth() - CIVal.countLeadingZeros()))
1706 return SE.getAddExpr(LHS, getSCEV(U->getOperand(1)));
1709 case Instruction::Xor:
1710 if (ConstantInt *CI = dyn_cast<ConstantInt>(U->getOperand(1))) {
1711 // If the RHS of the xor is a signbit, then this is just an add.
1712 // Instcombine turns add of signbit into xor as a strength reduction step.
1713 if (CI->getValue().isSignBit())
1714 return SE.getAddExpr(getSCEV(U->getOperand(0)),
1715 getSCEV(U->getOperand(1)));
1717 // If the RHS of xor is -1, then this is a not operation.
1718 else if (CI->isAllOnesValue())
1719 return SE.getNotSCEV(getSCEV(U->getOperand(0)));
1723 case Instruction::Shl:
1724 // Turn shift left of a constant amount into a multiply.
1725 if (ConstantInt *SA = dyn_cast<ConstantInt>(U->getOperand(1))) {
1726 uint32_t BitWidth = cast<IntegerType>(V->getType())->getBitWidth();
1727 Constant *X = ConstantInt::get(
1728 APInt(BitWidth, 1).shl(SA->getLimitedValue(BitWidth)));
1729 return SE.getMulExpr(getSCEV(U->getOperand(0)), getSCEV(X));
1733 case Instruction::LShr:
1734 // Turn logical shift right of a constant into a unsigned divide.
1735 if (ConstantInt *SA = dyn_cast<ConstantInt>(U->getOperand(1))) {
1736 uint32_t BitWidth = cast<IntegerType>(V->getType())->getBitWidth();
1737 Constant *X = ConstantInt::get(
1738 APInt(BitWidth, 1).shl(SA->getLimitedValue(BitWidth)));
1739 return SE.getUDivExpr(getSCEV(U->getOperand(0)), getSCEV(X));
1743 case Instruction::Trunc:
1744 return SE.getTruncateExpr(getSCEV(U->getOperand(0)), U->getType());
1746 case Instruction::ZExt:
1747 return SE.getZeroExtendExpr(getSCEV(U->getOperand(0)), U->getType());
1749 case Instruction::SExt:
1750 return SE.getSignExtendExpr(getSCEV(U->getOperand(0)), U->getType());
1752 case Instruction::BitCast:
1753 // BitCasts are no-op casts so we just eliminate the cast.
1754 if (U->getType()->isInteger() &&
1755 U->getOperand(0)->getType()->isInteger())
1756 return getSCEV(U->getOperand(0));
1759 case Instruction::PHI:
1760 return createNodeForPHI(cast<PHINode>(U));
1762 case Instruction::Select:
1763 // This could be a smax or umax that was lowered earlier.
1764 // Try to recover it.
1765 if (ICmpInst *ICI = dyn_cast<ICmpInst>(U->getOperand(0))) {
1766 Value *LHS = ICI->getOperand(0);
1767 Value *RHS = ICI->getOperand(1);
1768 switch (ICI->getPredicate()) {
1769 case ICmpInst::ICMP_SLT:
1770 case ICmpInst::ICMP_SLE:
1771 std::swap(LHS, RHS);
1773 case ICmpInst::ICMP_SGT:
1774 case ICmpInst::ICMP_SGE:
1775 if (LHS == U->getOperand(1) && RHS == U->getOperand(2))
1776 return SE.getSMaxExpr(getSCEV(LHS), getSCEV(RHS));
1777 else if (LHS == U->getOperand(2) && RHS == U->getOperand(1))
1778 // -smax(-x, -y) == smin(x, y).
1779 return SE.getNegativeSCEV(SE.getSMaxExpr(
1780 SE.getNegativeSCEV(getSCEV(LHS)),
1781 SE.getNegativeSCEV(getSCEV(RHS))));
1783 case ICmpInst::ICMP_ULT:
1784 case ICmpInst::ICMP_ULE:
1785 std::swap(LHS, RHS);
1787 case ICmpInst::ICMP_UGT:
1788 case ICmpInst::ICMP_UGE:
1789 if (LHS == U->getOperand(1) && RHS == U->getOperand(2))
1790 return SE.getUMaxExpr(getSCEV(LHS), getSCEV(RHS));
1791 else if (LHS == U->getOperand(2) && RHS == U->getOperand(1))
1792 // ~umax(~x, ~y) == umin(x, y)
1793 return SE.getNotSCEV(SE.getUMaxExpr(SE.getNotSCEV(getSCEV(LHS)),
1794 SE.getNotSCEV(getSCEV(RHS))));
1801 default: // We cannot analyze this expression.
1805 return SE.getUnknown(V);
1810 //===----------------------------------------------------------------------===//
1811 // Iteration Count Computation Code
1814 /// getIterationCount - If the specified loop has a predictable iteration
1815 /// count, return it. Note that it is not valid to call this method on a
1816 /// loop without a loop-invariant iteration count.
1817 SCEVHandle ScalarEvolutionsImpl::getIterationCount(const Loop *L) {
1818 std::map<const Loop*, SCEVHandle>::iterator I = IterationCounts.find(L);
1819 if (I == IterationCounts.end()) {
1820 SCEVHandle ItCount = ComputeIterationCount(L);
1821 I = IterationCounts.insert(std::make_pair(L, ItCount)).first;
1822 if (ItCount != UnknownValue) {
1823 assert(ItCount->isLoopInvariant(L) &&
1824 "Computed trip count isn't loop invariant for loop!");
1825 ++NumTripCountsComputed;
1826 } else if (isa<PHINode>(L->getHeader()->begin())) {
1827 // Only count loops that have phi nodes as not being computable.
1828 ++NumTripCountsNotComputed;
1834 /// ComputeIterationCount - Compute the number of times the specified loop
1836 SCEVHandle ScalarEvolutionsImpl::ComputeIterationCount(const Loop *L) {
1837 // If the loop has a non-one exit block count, we can't analyze it.
1838 SmallVector<BasicBlock*, 8> ExitBlocks;
1839 L->getExitBlocks(ExitBlocks);
1840 if (ExitBlocks.size() != 1) return UnknownValue;
1842 // Okay, there is one exit block. Try to find the condition that causes the
1843 // loop to be exited.
1844 BasicBlock *ExitBlock = ExitBlocks[0];
1846 BasicBlock *ExitingBlock = 0;
1847 for (pred_iterator PI = pred_begin(ExitBlock), E = pred_end(ExitBlock);
1849 if (L->contains(*PI)) {
1850 if (ExitingBlock == 0)
1853 return UnknownValue; // More than one block exiting!
1855 assert(ExitingBlock && "No exits from loop, something is broken!");
1857 // Okay, we've computed the exiting block. See what condition causes us to
1860 // FIXME: we should be able to handle switch instructions (with a single exit)
1861 BranchInst *ExitBr = dyn_cast<BranchInst>(ExitingBlock->getTerminator());
1862 if (ExitBr == 0) return UnknownValue;
1863 assert(ExitBr->isConditional() && "If unconditional, it can't be in loop!");
1865 // At this point, we know we have a conditional branch that determines whether
1866 // the loop is exited. However, we don't know if the branch is executed each
1867 // time through the loop. If not, then the execution count of the branch will
1868 // not be equal to the trip count of the loop.
1870 // Currently we check for this by checking to see if the Exit branch goes to
1871 // the loop header. If so, we know it will always execute the same number of
1872 // times as the loop. We also handle the case where the exit block *is* the
1873 // loop header. This is common for un-rotated loops. More extensive analysis
1874 // could be done to handle more cases here.
1875 if (ExitBr->getSuccessor(0) != L->getHeader() &&
1876 ExitBr->getSuccessor(1) != L->getHeader() &&
1877 ExitBr->getParent() != L->getHeader())
1878 return UnknownValue;
1880 ICmpInst *ExitCond = dyn_cast<ICmpInst>(ExitBr->getCondition());
1882 // If it's not an integer comparison then compute it the hard way.
1883 // Note that ICmpInst deals with pointer comparisons too so we must check
1884 // the type of the operand.
1885 if (ExitCond == 0 || isa<PointerType>(ExitCond->getOperand(0)->getType()))
1886 return ComputeIterationCountExhaustively(L, ExitBr->getCondition(),
1887 ExitBr->getSuccessor(0) == ExitBlock);
1889 // If the condition was exit on true, convert the condition to exit on false
1890 ICmpInst::Predicate Cond;
1891 if (ExitBr->getSuccessor(1) == ExitBlock)
1892 Cond = ExitCond->getPredicate();
1894 Cond = ExitCond->getInversePredicate();
1896 // Handle common loops like: for (X = "string"; *X; ++X)
1897 if (LoadInst *LI = dyn_cast<LoadInst>(ExitCond->getOperand(0)))
1898 if (Constant *RHS = dyn_cast<Constant>(ExitCond->getOperand(1))) {
1900 ComputeLoadConstantCompareIterationCount(LI, RHS, L, Cond);
1901 if (!isa<SCEVCouldNotCompute>(ItCnt)) return ItCnt;
1904 SCEVHandle LHS = getSCEV(ExitCond->getOperand(0));
1905 SCEVHandle RHS = getSCEV(ExitCond->getOperand(1));
1907 // Try to evaluate any dependencies out of the loop.
1908 SCEVHandle Tmp = getSCEVAtScope(LHS, L);
1909 if (!isa<SCEVCouldNotCompute>(Tmp)) LHS = Tmp;
1910 Tmp = getSCEVAtScope(RHS, L);
1911 if (!isa<SCEVCouldNotCompute>(Tmp)) RHS = Tmp;
1913 // At this point, we would like to compute how many iterations of the
1914 // loop the predicate will return true for these inputs.
1915 if (isa<SCEVConstant>(LHS) && !isa<SCEVConstant>(RHS)) {
1916 // If there is a constant, force it into the RHS.
1917 std::swap(LHS, RHS);
1918 Cond = ICmpInst::getSwappedPredicate(Cond);
1921 // FIXME: think about handling pointer comparisons! i.e.:
1922 // while (P != P+100) ++P;
1924 // If we have a comparison of a chrec against a constant, try to use value
1925 // ranges to answer this query.
1926 if (SCEVConstant *RHSC = dyn_cast<SCEVConstant>(RHS))
1927 if (SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(LHS))
1928 if (AddRec->getLoop() == L) {
1929 // Form the comparison range using the constant of the correct type so
1930 // that the ConstantRange class knows to do a signed or unsigned
1932 ConstantInt *CompVal = RHSC->getValue();
1933 const Type *RealTy = ExitCond->getOperand(0)->getType();
1934 CompVal = dyn_cast<ConstantInt>(
1935 ConstantExpr::getBitCast(CompVal, RealTy));
1937 // Form the constant range.
1938 ConstantRange CompRange(
1939 ICmpInst::makeConstantRange(Cond, CompVal->getValue()));
1941 SCEVHandle Ret = AddRec->getNumIterationsInRange(CompRange, SE);
1942 if (!isa<SCEVCouldNotCompute>(Ret)) return Ret;
1947 case ICmpInst::ICMP_NE: { // while (X != Y)
1948 // Convert to: while (X-Y != 0)
1949 SCEVHandle TC = HowFarToZero(SE.getMinusSCEV(LHS, RHS), L);
1950 if (!isa<SCEVCouldNotCompute>(TC)) return TC;
1953 case ICmpInst::ICMP_EQ: {
1954 // Convert to: while (X-Y == 0) // while (X == Y)
1955 SCEVHandle TC = HowFarToNonZero(SE.getMinusSCEV(LHS, RHS), L);
1956 if (!isa<SCEVCouldNotCompute>(TC)) return TC;
1959 case ICmpInst::ICMP_SLT: {
1960 SCEVHandle TC = HowManyLessThans(LHS, RHS, L, true);
1961 if (!isa<SCEVCouldNotCompute>(TC)) return TC;
1964 case ICmpInst::ICMP_SGT: {
1965 SCEVHandle TC = HowManyLessThans(SE.getNegativeSCEV(LHS),
1966 SE.getNegativeSCEV(RHS), L, true);
1967 if (!isa<SCEVCouldNotCompute>(TC)) return TC;
1970 case ICmpInst::ICMP_ULT: {
1971 SCEVHandle TC = HowManyLessThans(LHS, RHS, L, false);
1972 if (!isa<SCEVCouldNotCompute>(TC)) return TC;
1975 case ICmpInst::ICMP_UGT: {
1976 SCEVHandle TC = HowManyLessThans(SE.getNotSCEV(LHS),
1977 SE.getNotSCEV(RHS), L, false);
1978 if (!isa<SCEVCouldNotCompute>(TC)) return TC;
1983 cerr << "ComputeIterationCount ";
1984 if (ExitCond->getOperand(0)->getType()->isUnsigned())
1985 cerr << "[unsigned] ";
1987 << Instruction::getOpcodeName(Instruction::ICmp)
1988 << " " << *RHS << "\n";
1992 return ComputeIterationCountExhaustively(L, ExitCond,
1993 ExitBr->getSuccessor(0) == ExitBlock);
1996 static ConstantInt *
1997 EvaluateConstantChrecAtConstant(const SCEVAddRecExpr *AddRec, ConstantInt *C,
1998 ScalarEvolution &SE) {
1999 SCEVHandle InVal = SE.getConstant(C);
2000 SCEVHandle Val = AddRec->evaluateAtIteration(InVal, SE);
2001 assert(isa<SCEVConstant>(Val) &&
2002 "Evaluation of SCEV at constant didn't fold correctly?");
2003 return cast<SCEVConstant>(Val)->getValue();
2006 /// GetAddressedElementFromGlobal - Given a global variable with an initializer
2007 /// and a GEP expression (missing the pointer index) indexing into it, return
2008 /// the addressed element of the initializer or null if the index expression is
2011 GetAddressedElementFromGlobal(GlobalVariable *GV,
2012 const std::vector<ConstantInt*> &Indices) {
2013 Constant *Init = GV->getInitializer();
2014 for (unsigned i = 0, e = Indices.size(); i != e; ++i) {
2015 uint64_t Idx = Indices[i]->getZExtValue();
2016 if (ConstantStruct *CS = dyn_cast<ConstantStruct>(Init)) {
2017 assert(Idx < CS->getNumOperands() && "Bad struct index!");
2018 Init = cast<Constant>(CS->getOperand(Idx));
2019 } else if (ConstantArray *CA = dyn_cast<ConstantArray>(Init)) {
2020 if (Idx >= CA->getNumOperands()) return 0; // Bogus program
2021 Init = cast<Constant>(CA->getOperand(Idx));
2022 } else if (isa<ConstantAggregateZero>(Init)) {
2023 if (const StructType *STy = dyn_cast<StructType>(Init->getType())) {
2024 assert(Idx < STy->getNumElements() && "Bad struct index!");
2025 Init = Constant::getNullValue(STy->getElementType(Idx));
2026 } else if (const ArrayType *ATy = dyn_cast<ArrayType>(Init->getType())) {
2027 if (Idx >= ATy->getNumElements()) return 0; // Bogus program
2028 Init = Constant::getNullValue(ATy->getElementType());
2030 assert(0 && "Unknown constant aggregate type!");
2034 return 0; // Unknown initializer type
2040 /// ComputeLoadConstantCompareIterationCount - Given an exit condition of
2041 /// 'icmp op load X, cst', try to see if we can compute the trip count.
2042 SCEVHandle ScalarEvolutionsImpl::
2043 ComputeLoadConstantCompareIterationCount(LoadInst *LI, Constant *RHS,
2045 ICmpInst::Predicate predicate) {
2046 if (LI->isVolatile()) return UnknownValue;
2048 // Check to see if the loaded pointer is a getelementptr of a global.
2049 GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(LI->getOperand(0));
2050 if (!GEP) return UnknownValue;
2052 // Make sure that it is really a constant global we are gepping, with an
2053 // initializer, and make sure the first IDX is really 0.
2054 GlobalVariable *GV = dyn_cast<GlobalVariable>(GEP->getOperand(0));
2055 if (!GV || !GV->isConstant() || !GV->hasInitializer() ||
2056 GEP->getNumOperands() < 3 || !isa<Constant>(GEP->getOperand(1)) ||
2057 !cast<Constant>(GEP->getOperand(1))->isNullValue())
2058 return UnknownValue;
2060 // Okay, we allow one non-constant index into the GEP instruction.
2062 std::vector<ConstantInt*> Indexes;
2063 unsigned VarIdxNum = 0;
2064 for (unsigned i = 2, e = GEP->getNumOperands(); i != e; ++i)
2065 if (ConstantInt *CI = dyn_cast<ConstantInt>(GEP->getOperand(i))) {
2066 Indexes.push_back(CI);
2067 } else if (!isa<ConstantInt>(GEP->getOperand(i))) {
2068 if (VarIdx) return UnknownValue; // Multiple non-constant idx's.
2069 VarIdx = GEP->getOperand(i);
2071 Indexes.push_back(0);
2074 // Okay, we know we have a (load (gep GV, 0, X)) comparison with a constant.
2075 // Check to see if X is a loop variant variable value now.
2076 SCEVHandle Idx = getSCEV(VarIdx);
2077 SCEVHandle Tmp = getSCEVAtScope(Idx, L);
2078 if (!isa<SCEVCouldNotCompute>(Tmp)) Idx = Tmp;
2080 // We can only recognize very limited forms of loop index expressions, in
2081 // particular, only affine AddRec's like {C1,+,C2}.
2082 SCEVAddRecExpr *IdxExpr = dyn_cast<SCEVAddRecExpr>(Idx);
2083 if (!IdxExpr || !IdxExpr->isAffine() || IdxExpr->isLoopInvariant(L) ||
2084 !isa<SCEVConstant>(IdxExpr->getOperand(0)) ||
2085 !isa<SCEVConstant>(IdxExpr->getOperand(1)))
2086 return UnknownValue;
2088 unsigned MaxSteps = MaxBruteForceIterations;
2089 for (unsigned IterationNum = 0; IterationNum != MaxSteps; ++IterationNum) {
2090 ConstantInt *ItCst =
2091 ConstantInt::get(IdxExpr->getType(), IterationNum);
2092 ConstantInt *Val = EvaluateConstantChrecAtConstant(IdxExpr, ItCst, SE);
2094 // Form the GEP offset.
2095 Indexes[VarIdxNum] = Val;
2097 Constant *Result = GetAddressedElementFromGlobal(GV, Indexes);
2098 if (Result == 0) break; // Cannot compute!
2100 // Evaluate the condition for this iteration.
2101 Result = ConstantExpr::getICmp(predicate, Result, RHS);
2102 if (!isa<ConstantInt>(Result)) break; // Couldn't decide for sure
2103 if (cast<ConstantInt>(Result)->getValue().isMinValue()) {
2105 cerr << "\n***\n*** Computed loop count " << *ItCst
2106 << "\n*** From global " << *GV << "*** BB: " << *L->getHeader()
2109 ++NumArrayLenItCounts;
2110 return SE.getConstant(ItCst); // Found terminating iteration!
2113 return UnknownValue;
2117 /// CanConstantFold - Return true if we can constant fold an instruction of the
2118 /// specified type, assuming that all operands were constants.
2119 static bool CanConstantFold(const Instruction *I) {
2120 if (isa<BinaryOperator>(I) || isa<CmpInst>(I) ||
2121 isa<SelectInst>(I) || isa<CastInst>(I) || isa<GetElementPtrInst>(I))
2124 if (const CallInst *CI = dyn_cast<CallInst>(I))
2125 if (const Function *F = CI->getCalledFunction())
2126 return canConstantFoldCallTo(F);
2130 /// getConstantEvolvingPHI - Given an LLVM value and a loop, return a PHI node
2131 /// in the loop that V is derived from. We allow arbitrary operations along the
2132 /// way, but the operands of an operation must either be constants or a value
2133 /// derived from a constant PHI. If this expression does not fit with these
2134 /// constraints, return null.
2135 static PHINode *getConstantEvolvingPHI(Value *V, const Loop *L) {
2136 // If this is not an instruction, or if this is an instruction outside of the
2137 // loop, it can't be derived from a loop PHI.
2138 Instruction *I = dyn_cast<Instruction>(V);
2139 if (I == 0 || !L->contains(I->getParent())) return 0;
2141 if (PHINode *PN = dyn_cast<PHINode>(I)) {
2142 if (L->getHeader() == I->getParent())
2145 // We don't currently keep track of the control flow needed to evaluate
2146 // PHIs, so we cannot handle PHIs inside of loops.
2150 // If we won't be able to constant fold this expression even if the operands
2151 // are constants, return early.
2152 if (!CanConstantFold(I)) return 0;
2154 // Otherwise, we can evaluate this instruction if all of its operands are
2155 // constant or derived from a PHI node themselves.
2157 for (unsigned Op = 0, e = I->getNumOperands(); Op != e; ++Op)
2158 if (!(isa<Constant>(I->getOperand(Op)) ||
2159 isa<GlobalValue>(I->getOperand(Op)))) {
2160 PHINode *P = getConstantEvolvingPHI(I->getOperand(Op), L);
2161 if (P == 0) return 0; // Not evolving from PHI
2165 return 0; // Evolving from multiple different PHIs.
2168 // This is a expression evolving from a constant PHI!
2172 /// EvaluateExpression - Given an expression that passes the
2173 /// getConstantEvolvingPHI predicate, evaluate its value assuming the PHI node
2174 /// in the loop has the value PHIVal. If we can't fold this expression for some
2175 /// reason, return null.
2176 static Constant *EvaluateExpression(Value *V, Constant *PHIVal) {
2177 if (isa<PHINode>(V)) return PHIVal;
2178 if (Constant *C = dyn_cast<Constant>(V)) return C;
2179 Instruction *I = cast<Instruction>(V);
2181 std::vector<Constant*> Operands;
2182 Operands.resize(I->getNumOperands());
2184 for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i) {
2185 Operands[i] = EvaluateExpression(I->getOperand(i), PHIVal);
2186 if (Operands[i] == 0) return 0;
2189 if (const CmpInst *CI = dyn_cast<CmpInst>(I))
2190 return ConstantFoldCompareInstOperands(CI->getPredicate(),
2191 &Operands[0], Operands.size());
2193 return ConstantFoldInstOperands(I->getOpcode(), I->getType(),
2194 &Operands[0], Operands.size());
2197 /// getConstantEvolutionLoopExitValue - If we know that the specified Phi is
2198 /// in the header of its containing loop, we know the loop executes a
2199 /// constant number of times, and the PHI node is just a recurrence
2200 /// involving constants, fold it.
2201 Constant *ScalarEvolutionsImpl::
2202 getConstantEvolutionLoopExitValue(PHINode *PN, const APInt& Its, const Loop *L){
2203 std::map<PHINode*, Constant*>::iterator I =
2204 ConstantEvolutionLoopExitValue.find(PN);
2205 if (I != ConstantEvolutionLoopExitValue.end())
2208 if (Its.ugt(APInt(Its.getBitWidth(),MaxBruteForceIterations)))
2209 return ConstantEvolutionLoopExitValue[PN] = 0; // Not going to evaluate it.
2211 Constant *&RetVal = ConstantEvolutionLoopExitValue[PN];
2213 // Since the loop is canonicalized, the PHI node must have two entries. One
2214 // entry must be a constant (coming in from outside of the loop), and the
2215 // second must be derived from the same PHI.
2216 bool SecondIsBackedge = L->contains(PN->getIncomingBlock(1));
2217 Constant *StartCST =
2218 dyn_cast<Constant>(PN->getIncomingValue(!SecondIsBackedge));
2220 return RetVal = 0; // Must be a constant.
2222 Value *BEValue = PN->getIncomingValue(SecondIsBackedge);
2223 PHINode *PN2 = getConstantEvolvingPHI(BEValue, L);
2225 return RetVal = 0; // Not derived from same PHI.
2227 // Execute the loop symbolically to determine the exit value.
2228 if (Its.getActiveBits() >= 32)
2229 return RetVal = 0; // More than 2^32-1 iterations?? Not doing it!
2231 unsigned NumIterations = Its.getZExtValue(); // must be in range
2232 unsigned IterationNum = 0;
2233 for (Constant *PHIVal = StartCST; ; ++IterationNum) {
2234 if (IterationNum == NumIterations)
2235 return RetVal = PHIVal; // Got exit value!
2237 // Compute the value of the PHI node for the next iteration.
2238 Constant *NextPHI = EvaluateExpression(BEValue, PHIVal);
2239 if (NextPHI == PHIVal)
2240 return RetVal = NextPHI; // Stopped evolving!
2242 return 0; // Couldn't evaluate!
2247 /// ComputeIterationCountExhaustively - If the trip is known to execute a
2248 /// constant number of times (the condition evolves only from constants),
2249 /// try to evaluate a few iterations of the loop until we get the exit
2250 /// condition gets a value of ExitWhen (true or false). If we cannot
2251 /// evaluate the trip count of the loop, return UnknownValue.
2252 SCEVHandle ScalarEvolutionsImpl::
2253 ComputeIterationCountExhaustively(const Loop *L, Value *Cond, bool ExitWhen) {
2254 PHINode *PN = getConstantEvolvingPHI(Cond, L);
2255 if (PN == 0) return UnknownValue;
2257 // Since the loop is canonicalized, the PHI node must have two entries. One
2258 // entry must be a constant (coming in from outside of the loop), and the
2259 // second must be derived from the same PHI.
2260 bool SecondIsBackedge = L->contains(PN->getIncomingBlock(1));
2261 Constant *StartCST =
2262 dyn_cast<Constant>(PN->getIncomingValue(!SecondIsBackedge));
2263 if (StartCST == 0) return UnknownValue; // Must be a constant.
2265 Value *BEValue = PN->getIncomingValue(SecondIsBackedge);
2266 PHINode *PN2 = getConstantEvolvingPHI(BEValue, L);
2267 if (PN2 != PN) return UnknownValue; // Not derived from same PHI.
2269 // Okay, we find a PHI node that defines the trip count of this loop. Execute
2270 // the loop symbolically to determine when the condition gets a value of
2272 unsigned IterationNum = 0;
2273 unsigned MaxIterations = MaxBruteForceIterations; // Limit analysis.
2274 for (Constant *PHIVal = StartCST;
2275 IterationNum != MaxIterations; ++IterationNum) {
2276 ConstantInt *CondVal =
2277 dyn_cast_or_null<ConstantInt>(EvaluateExpression(Cond, PHIVal));
2279 // Couldn't symbolically evaluate.
2280 if (!CondVal) return UnknownValue;
2282 if (CondVal->getValue() == uint64_t(ExitWhen)) {
2283 ConstantEvolutionLoopExitValue[PN] = PHIVal;
2284 ++NumBruteForceTripCountsComputed;
2285 return SE.getConstant(ConstantInt::get(Type::Int32Ty, IterationNum));
2288 // Compute the value of the PHI node for the next iteration.
2289 Constant *NextPHI = EvaluateExpression(BEValue, PHIVal);
2290 if (NextPHI == 0 || NextPHI == PHIVal)
2291 return UnknownValue; // Couldn't evaluate or not making progress...
2295 // Too many iterations were needed to evaluate.
2296 return UnknownValue;
2299 /// getSCEVAtScope - Compute the value of the specified expression within the
2300 /// indicated loop (which may be null to indicate in no loop). If the
2301 /// expression cannot be evaluated, return UnknownValue.
2302 SCEVHandle ScalarEvolutionsImpl::getSCEVAtScope(SCEV *V, const Loop *L) {
2303 // FIXME: this should be turned into a virtual method on SCEV!
2305 if (isa<SCEVConstant>(V)) return V;
2307 // If this instruction is evolved from a constant-evolving PHI, compute the
2308 // exit value from the loop without using SCEVs.
2309 if (SCEVUnknown *SU = dyn_cast<SCEVUnknown>(V)) {
2310 if (Instruction *I = dyn_cast<Instruction>(SU->getValue())) {
2311 const Loop *LI = this->LI[I->getParent()];
2312 if (LI && LI->getParentLoop() == L) // Looking for loop exit value.
2313 if (PHINode *PN = dyn_cast<PHINode>(I))
2314 if (PN->getParent() == LI->getHeader()) {
2315 // Okay, there is no closed form solution for the PHI node. Check
2316 // to see if the loop that contains it has a known iteration count.
2317 // If so, we may be able to force computation of the exit value.
2318 SCEVHandle IterationCount = getIterationCount(LI);
2319 if (SCEVConstant *ICC = dyn_cast<SCEVConstant>(IterationCount)) {
2320 // Okay, we know how many times the containing loop executes. If
2321 // this is a constant evolving PHI node, get the final value at
2322 // the specified iteration number.
2323 Constant *RV = getConstantEvolutionLoopExitValue(PN,
2324 ICC->getValue()->getValue(),
2326 if (RV) return SE.getUnknown(RV);
2330 // Okay, this is an expression that we cannot symbolically evaluate
2331 // into a SCEV. Check to see if it's possible to symbolically evaluate
2332 // the arguments into constants, and if so, try to constant propagate the
2333 // result. This is particularly useful for computing loop exit values.
2334 if (CanConstantFold(I)) {
2335 std::vector<Constant*> Operands;
2336 Operands.reserve(I->getNumOperands());
2337 for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i) {
2338 Value *Op = I->getOperand(i);
2339 if (Constant *C = dyn_cast<Constant>(Op)) {
2340 Operands.push_back(C);
2342 // If any of the operands is non-constant and if they are
2343 // non-integer, don't even try to analyze them with scev techniques.
2344 if (!isa<IntegerType>(Op->getType()))
2347 SCEVHandle OpV = getSCEVAtScope(getSCEV(Op), L);
2348 if (SCEVConstant *SC = dyn_cast<SCEVConstant>(OpV))
2349 Operands.push_back(ConstantExpr::getIntegerCast(SC->getValue(),
2352 else if (SCEVUnknown *SU = dyn_cast<SCEVUnknown>(OpV)) {
2353 if (Constant *C = dyn_cast<Constant>(SU->getValue()))
2354 Operands.push_back(ConstantExpr::getIntegerCast(C,
2366 if (const CmpInst *CI = dyn_cast<CmpInst>(I))
2367 C = ConstantFoldCompareInstOperands(CI->getPredicate(),
2368 &Operands[0], Operands.size());
2370 C = ConstantFoldInstOperands(I->getOpcode(), I->getType(),
2371 &Operands[0], Operands.size());
2372 return SE.getUnknown(C);
2376 // This is some other type of SCEVUnknown, just return it.
2380 if (SCEVCommutativeExpr *Comm = dyn_cast<SCEVCommutativeExpr>(V)) {
2381 // Avoid performing the look-up in the common case where the specified
2382 // expression has no loop-variant portions.
2383 for (unsigned i = 0, e = Comm->getNumOperands(); i != e; ++i) {
2384 SCEVHandle OpAtScope = getSCEVAtScope(Comm->getOperand(i), L);
2385 if (OpAtScope != Comm->getOperand(i)) {
2386 if (OpAtScope == UnknownValue) return UnknownValue;
2387 // Okay, at least one of these operands is loop variant but might be
2388 // foldable. Build a new instance of the folded commutative expression.
2389 std::vector<SCEVHandle> NewOps(Comm->op_begin(), Comm->op_begin()+i);
2390 NewOps.push_back(OpAtScope);
2392 for (++i; i != e; ++i) {
2393 OpAtScope = getSCEVAtScope(Comm->getOperand(i), L);
2394 if (OpAtScope == UnknownValue) return UnknownValue;
2395 NewOps.push_back(OpAtScope);
2397 if (isa<SCEVAddExpr>(Comm))
2398 return SE.getAddExpr(NewOps);
2399 if (isa<SCEVMulExpr>(Comm))
2400 return SE.getMulExpr(NewOps);
2401 if (isa<SCEVSMaxExpr>(Comm))
2402 return SE.getSMaxExpr(NewOps);
2403 if (isa<SCEVUMaxExpr>(Comm))
2404 return SE.getUMaxExpr(NewOps);
2405 assert(0 && "Unknown commutative SCEV type!");
2408 // If we got here, all operands are loop invariant.
2412 if (SCEVUDivExpr *Div = dyn_cast<SCEVUDivExpr>(V)) {
2413 SCEVHandle LHS = getSCEVAtScope(Div->getLHS(), L);
2414 if (LHS == UnknownValue) return LHS;
2415 SCEVHandle RHS = getSCEVAtScope(Div->getRHS(), L);
2416 if (RHS == UnknownValue) return RHS;
2417 if (LHS == Div->getLHS() && RHS == Div->getRHS())
2418 return Div; // must be loop invariant
2419 return SE.getUDivExpr(LHS, RHS);
2422 // If this is a loop recurrence for a loop that does not contain L, then we
2423 // are dealing with the final value computed by the loop.
2424 if (SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(V)) {
2425 if (!L || !AddRec->getLoop()->contains(L->getHeader())) {
2426 // To evaluate this recurrence, we need to know how many times the AddRec
2427 // loop iterates. Compute this now.
2428 SCEVHandle IterationCount = getIterationCount(AddRec->getLoop());
2429 if (IterationCount == UnknownValue) return UnknownValue;
2430 IterationCount = SE.getTruncateOrZeroExtend(IterationCount,
2433 // If the value is affine, simplify the expression evaluation to just
2434 // Start + Step*IterationCount.
2435 if (AddRec->isAffine())
2436 return SE.getAddExpr(AddRec->getStart(),
2437 SE.getMulExpr(IterationCount,
2438 AddRec->getOperand(1)));
2440 // Otherwise, evaluate it the hard way.
2441 return AddRec->evaluateAtIteration(IterationCount, SE);
2443 return UnknownValue;
2446 //assert(0 && "Unknown SCEV type!");
2447 return UnknownValue;
2450 /// SolveLinEquationWithOverflow - Finds the minimum unsigned root of the
2451 /// following equation:
2453 /// A * X = B (mod N)
2455 /// where N = 2^BW and BW is the common bit width of A and B. The signedness of
2456 /// A and B isn't important.
2458 /// If the equation does not have a solution, SCEVCouldNotCompute is returned.
2459 static SCEVHandle SolveLinEquationWithOverflow(const APInt &A, const APInt &B,
2460 ScalarEvolution &SE) {
2461 uint32_t BW = A.getBitWidth();
2462 assert(BW == B.getBitWidth() && "Bit widths must be the same.");
2463 assert(A != 0 && "A must be non-zero.");
2467 // The gcd of A and N may have only one prime factor: 2. The number of
2468 // trailing zeros in A is its multiplicity
2469 uint32_t Mult2 = A.countTrailingZeros();
2472 // 2. Check if B is divisible by D.
2474 // B is divisible by D if and only if the multiplicity of prime factor 2 for B
2475 // is not less than multiplicity of this prime factor for D.
2476 if (B.countTrailingZeros() < Mult2)
2477 return new SCEVCouldNotCompute();
2479 // 3. Compute I: the multiplicative inverse of (A / D) in arithmetic
2482 // (N / D) may need BW+1 bits in its representation. Hence, we'll use this
2483 // bit width during computations.
2484 APInt AD = A.lshr(Mult2).zext(BW + 1); // AD = A / D
2485 APInt Mod(BW + 1, 0);
2486 Mod.set(BW - Mult2); // Mod = N / D
2487 APInt I = AD.multiplicativeInverse(Mod);
2489 // 4. Compute the minimum unsigned root of the equation:
2490 // I * (B / D) mod (N / D)
2491 APInt Result = (I * B.lshr(Mult2).zext(BW + 1)).urem(Mod);
2493 // The result is guaranteed to be less than 2^BW so we may truncate it to BW
2495 return SE.getConstant(Result.trunc(BW));
2498 /// SolveQuadraticEquation - Find the roots of the quadratic equation for the
2499 /// given quadratic chrec {L,+,M,+,N}. This returns either the two roots (which
2500 /// might be the same) or two SCEVCouldNotCompute objects.
2502 static std::pair<SCEVHandle,SCEVHandle>
2503 SolveQuadraticEquation(const SCEVAddRecExpr *AddRec, ScalarEvolution &SE) {
2504 assert(AddRec->getNumOperands() == 3 && "This is not a quadratic chrec!");
2505 SCEVConstant *LC = dyn_cast<SCEVConstant>(AddRec->getOperand(0));
2506 SCEVConstant *MC = dyn_cast<SCEVConstant>(AddRec->getOperand(1));
2507 SCEVConstant *NC = dyn_cast<SCEVConstant>(AddRec->getOperand(2));
2509 // We currently can only solve this if the coefficients are constants.
2510 if (!LC || !MC || !NC) {
2511 SCEV *CNC = new SCEVCouldNotCompute();
2512 return std::make_pair(CNC, CNC);
2515 uint32_t BitWidth = LC->getValue()->getValue().getBitWidth();
2516 const APInt &L = LC->getValue()->getValue();
2517 const APInt &M = MC->getValue()->getValue();
2518 const APInt &N = NC->getValue()->getValue();
2519 APInt Two(BitWidth, 2);
2520 APInt Four(BitWidth, 4);
2523 using namespace APIntOps;
2525 // Convert from chrec coefficients to polynomial coefficients AX^2+BX+C
2526 // The B coefficient is M-N/2
2530 // The A coefficient is N/2
2531 APInt A(N.sdiv(Two));
2533 // Compute the B^2-4ac term.
2536 SqrtTerm -= Four * (A * C);
2538 // Compute sqrt(B^2-4ac). This is guaranteed to be the nearest
2539 // integer value or else APInt::sqrt() will assert.
2540 APInt SqrtVal(SqrtTerm.sqrt());
2542 // Compute the two solutions for the quadratic formula.
2543 // The divisions must be performed as signed divisions.
2545 APInt TwoA( A << 1 );
2546 ConstantInt *Solution1 = ConstantInt::get((NegB + SqrtVal).sdiv(TwoA));
2547 ConstantInt *Solution2 = ConstantInt::get((NegB - SqrtVal).sdiv(TwoA));
2549 return std::make_pair(SE.getConstant(Solution1),
2550 SE.getConstant(Solution2));
2551 } // end APIntOps namespace
2554 /// HowFarToZero - Return the number of times a backedge comparing the specified
2555 /// value to zero will execute. If not computable, return UnknownValue
2556 SCEVHandle ScalarEvolutionsImpl::HowFarToZero(SCEV *V, const Loop *L) {
2557 // If the value is a constant
2558 if (SCEVConstant *C = dyn_cast<SCEVConstant>(V)) {
2559 // If the value is already zero, the branch will execute zero times.
2560 if (C->getValue()->isZero()) return C;
2561 return UnknownValue; // Otherwise it will loop infinitely.
2564 SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(V);
2565 if (!AddRec || AddRec->getLoop() != L)
2566 return UnknownValue;
2568 if (AddRec->isAffine()) {
2569 // If this is an affine expression, the execution count of this branch is
2570 // the minimum unsigned root of the following equation:
2572 // Start + Step*N = 0 (mod 2^BW)
2576 // Step*N = -Start (mod 2^BW)
2578 // where BW is the common bit width of Start and Step.
2580 // Get the initial value for the loop.
2581 SCEVHandle Start = getSCEVAtScope(AddRec->getStart(), L->getParentLoop());
2582 if (isa<SCEVCouldNotCompute>(Start)) return UnknownValue;
2584 SCEVHandle Step = getSCEVAtScope(AddRec->getOperand(1), L->getParentLoop());
2586 if (SCEVConstant *StepC = dyn_cast<SCEVConstant>(Step)) {
2587 // For now we handle only constant steps.
2589 // First, handle unitary steps.
2590 if (StepC->getValue()->equalsInt(1)) // 1*N = -Start (mod 2^BW), so:
2591 return SE.getNegativeSCEV(Start); // N = -Start (as unsigned)
2592 if (StepC->getValue()->isAllOnesValue()) // -1*N = -Start (mod 2^BW), so:
2593 return Start; // N = Start (as unsigned)
2595 // Then, try to solve the above equation provided that Start is constant.
2596 if (SCEVConstant *StartC = dyn_cast<SCEVConstant>(Start))
2597 return SolveLinEquationWithOverflow(StepC->getValue()->getValue(),
2598 -StartC->getValue()->getValue(),SE);
2600 } else if (AddRec->isQuadratic() && AddRec->getType()->isInteger()) {
2601 // If this is a quadratic (3-term) AddRec {L,+,M,+,N}, find the roots of
2602 // the quadratic equation to solve it.
2603 std::pair<SCEVHandle,SCEVHandle> Roots = SolveQuadraticEquation(AddRec, SE);
2604 SCEVConstant *R1 = dyn_cast<SCEVConstant>(Roots.first);
2605 SCEVConstant *R2 = dyn_cast<SCEVConstant>(Roots.second);
2608 cerr << "HFTZ: " << *V << " - sol#1: " << *R1
2609 << " sol#2: " << *R2 << "\n";
2611 // Pick the smallest positive root value.
2612 if (ConstantInt *CB =
2613 dyn_cast<ConstantInt>(ConstantExpr::getICmp(ICmpInst::ICMP_ULT,
2614 R1->getValue(), R2->getValue()))) {
2615 if (CB->getZExtValue() == false)
2616 std::swap(R1, R2); // R1 is the minimum root now.
2618 // We can only use this value if the chrec ends up with an exact zero
2619 // value at this index. When solving for "X*X != 5", for example, we
2620 // should not accept a root of 2.
2621 SCEVHandle Val = AddRec->evaluateAtIteration(R1, SE);
2623 return R1; // We found a quadratic root!
2628 return UnknownValue;
2631 /// HowFarToNonZero - Return the number of times a backedge checking the
2632 /// specified value for nonzero will execute. If not computable, return
2634 SCEVHandle ScalarEvolutionsImpl::HowFarToNonZero(SCEV *V, const Loop *L) {
2635 // Loops that look like: while (X == 0) are very strange indeed. We don't
2636 // handle them yet except for the trivial case. This could be expanded in the
2637 // future as needed.
2639 // If the value is a constant, check to see if it is known to be non-zero
2640 // already. If so, the backedge will execute zero times.
2641 if (SCEVConstant *C = dyn_cast<SCEVConstant>(V)) {
2642 if (!C->getValue()->isNullValue())
2643 return SE.getIntegerSCEV(0, C->getType());
2644 return UnknownValue; // Otherwise it will loop infinitely.
2647 // We could implement others, but I really doubt anyone writes loops like
2648 // this, and if they did, they would already be constant folded.
2649 return UnknownValue;
2652 /// executesAtLeastOnce - Test whether entry to the loop is protected by
2653 /// a conditional between LHS and RHS.
2654 bool ScalarEvolutionsImpl::executesAtLeastOnce(const Loop *L, bool isSigned,
2655 SCEV *LHS, SCEV *RHS) {
2656 BasicBlock *Preheader = L->getLoopPreheader();
2657 BasicBlock *PreheaderDest = L->getHeader();
2658 if (Preheader == 0) return false;
2660 BranchInst *LoopEntryPredicate =
2661 dyn_cast<BranchInst>(Preheader->getTerminator());
2662 if (!LoopEntryPredicate) return false;
2664 // This might be a critical edge broken out. If the loop preheader ends in
2665 // an unconditional branch to the loop, check to see if the preheader has a
2666 // single predecessor, and if so, look for its terminator.
2667 while (LoopEntryPredicate->isUnconditional()) {
2668 PreheaderDest = Preheader;
2669 Preheader = Preheader->getSinglePredecessor();
2670 if (!Preheader) return false; // Multiple preds.
2672 LoopEntryPredicate =
2673 dyn_cast<BranchInst>(Preheader->getTerminator());
2674 if (!LoopEntryPredicate) return false;
2677 ICmpInst *ICI = dyn_cast<ICmpInst>(LoopEntryPredicate->getCondition());
2678 if (!ICI) return false;
2680 // Now that we found a conditional branch that dominates the loop, check to
2681 // see if it is the comparison we are looking for.
2682 Value *PreCondLHS = ICI->getOperand(0);
2683 Value *PreCondRHS = ICI->getOperand(1);
2684 ICmpInst::Predicate Cond;
2685 if (LoopEntryPredicate->getSuccessor(0) == PreheaderDest)
2686 Cond = ICI->getPredicate();
2688 Cond = ICI->getInversePredicate();
2691 case ICmpInst::ICMP_UGT:
2692 if (isSigned) return false;
2693 std::swap(PreCondLHS, PreCondRHS);
2694 Cond = ICmpInst::ICMP_ULT;
2696 case ICmpInst::ICMP_SGT:
2697 if (!isSigned) return false;
2698 std::swap(PreCondLHS, PreCondRHS);
2699 Cond = ICmpInst::ICMP_SLT;
2701 case ICmpInst::ICMP_ULT:
2702 if (isSigned) return false;
2704 case ICmpInst::ICMP_SLT:
2705 if (!isSigned) return false;
2711 if (!PreCondLHS->getType()->isInteger()) return false;
2713 return LHS == getSCEV(PreCondLHS) && RHS == getSCEV(PreCondRHS);
2716 /// HowManyLessThans - Return the number of times a backedge containing the
2717 /// specified less-than comparison will execute. If not computable, return
2719 SCEVHandle ScalarEvolutionsImpl::
2720 HowManyLessThans(SCEV *LHS, SCEV *RHS, const Loop *L, bool isSigned) {
2721 // Only handle: "ADDREC < LoopInvariant".
2722 if (!RHS->isLoopInvariant(L)) return UnknownValue;
2724 SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(LHS);
2725 if (!AddRec || AddRec->getLoop() != L)
2726 return UnknownValue;
2728 if (AddRec->isAffine()) {
2729 // FORNOW: We only support unit strides.
2730 SCEVHandle One = SE.getIntegerSCEV(1, RHS->getType());
2731 if (AddRec->getOperand(1) != One)
2732 return UnknownValue;
2734 // We know the LHS is of the form {n,+,1} and the RHS is some loop-invariant
2735 // m. So, we count the number of iterations in which {n,+,1} < m is true.
2736 // Note that we cannot simply return max(m-n,0) because it's not safe to
2737 // treat m-n as signed nor unsigned due to overflow possibility.
2739 // First, we get the value of the LHS in the first iteration: n
2740 SCEVHandle Start = AddRec->getOperand(0);
2742 if (executesAtLeastOnce(L, isSigned,
2743 SE.getMinusSCEV(AddRec->getOperand(0), One), RHS)) {
2744 // Since we know that the condition is true in order to enter the loop,
2745 // we know that it will run exactly m-n times.
2746 return SE.getMinusSCEV(RHS, Start);
2748 // Then, we get the value of the LHS in the first iteration in which the
2749 // above condition doesn't hold. This equals to max(m,n).
2750 SCEVHandle End = isSigned ? SE.getSMaxExpr(RHS, Start)
2751 : SE.getUMaxExpr(RHS, Start);
2753 // Finally, we subtract these two values to get the number of times the
2754 // backedge is executed: max(m,n)-n.
2755 return SE.getMinusSCEV(End, Start);
2759 return UnknownValue;
2762 /// getNumIterationsInRange - Return the number of iterations of this loop that
2763 /// produce values in the specified constant range. Another way of looking at
2764 /// this is that it returns the first iteration number where the value is not in
2765 /// the condition, thus computing the exit count. If the iteration count can't
2766 /// be computed, an instance of SCEVCouldNotCompute is returned.
2767 SCEVHandle SCEVAddRecExpr::getNumIterationsInRange(ConstantRange Range,
2768 ScalarEvolution &SE) const {
2769 if (Range.isFullSet()) // Infinite loop.
2770 return new SCEVCouldNotCompute();
2772 // If the start is a non-zero constant, shift the range to simplify things.
2773 if (SCEVConstant *SC = dyn_cast<SCEVConstant>(getStart()))
2774 if (!SC->getValue()->isZero()) {
2775 std::vector<SCEVHandle> Operands(op_begin(), op_end());
2776 Operands[0] = SE.getIntegerSCEV(0, SC->getType());
2777 SCEVHandle Shifted = SE.getAddRecExpr(Operands, getLoop());
2778 if (SCEVAddRecExpr *ShiftedAddRec = dyn_cast<SCEVAddRecExpr>(Shifted))
2779 return ShiftedAddRec->getNumIterationsInRange(
2780 Range.subtract(SC->getValue()->getValue()), SE);
2781 // This is strange and shouldn't happen.
2782 return new SCEVCouldNotCompute();
2785 // The only time we can solve this is when we have all constant indices.
2786 // Otherwise, we cannot determine the overflow conditions.
2787 for (unsigned i = 0, e = getNumOperands(); i != e; ++i)
2788 if (!isa<SCEVConstant>(getOperand(i)))
2789 return new SCEVCouldNotCompute();
2792 // Okay at this point we know that all elements of the chrec are constants and
2793 // that the start element is zero.
2795 // First check to see if the range contains zero. If not, the first
2797 if (!Range.contains(APInt(getBitWidth(),0)))
2798 return SE.getConstant(ConstantInt::get(getType(),0));
2801 // If this is an affine expression then we have this situation:
2802 // Solve {0,+,A} in Range === Ax in Range
2804 // We know that zero is in the range. If A is positive then we know that
2805 // the upper value of the range must be the first possible exit value.
2806 // If A is negative then the lower of the range is the last possible loop
2807 // value. Also note that we already checked for a full range.
2808 APInt One(getBitWidth(),1);
2809 APInt A = cast<SCEVConstant>(getOperand(1))->getValue()->getValue();
2810 APInt End = A.sge(One) ? (Range.getUpper() - One) : Range.getLower();
2812 // The exit value should be (End+A)/A.
2813 APInt ExitVal = (End + A).udiv(A);
2814 ConstantInt *ExitValue = ConstantInt::get(ExitVal);
2816 // Evaluate at the exit value. If we really did fall out of the valid
2817 // range, then we computed our trip count, otherwise wrap around or other
2818 // things must have happened.
2819 ConstantInt *Val = EvaluateConstantChrecAtConstant(this, ExitValue, SE);
2820 if (Range.contains(Val->getValue()))
2821 return new SCEVCouldNotCompute(); // Something strange happened
2823 // Ensure that the previous value is in the range. This is a sanity check.
2824 assert(Range.contains(
2825 EvaluateConstantChrecAtConstant(this,
2826 ConstantInt::get(ExitVal - One), SE)->getValue()) &&
2827 "Linear scev computation is off in a bad way!");
2828 return SE.getConstant(ExitValue);
2829 } else if (isQuadratic()) {
2830 // If this is a quadratic (3-term) AddRec {L,+,M,+,N}, find the roots of the
2831 // quadratic equation to solve it. To do this, we must frame our problem in
2832 // terms of figuring out when zero is crossed, instead of when
2833 // Range.getUpper() is crossed.
2834 std::vector<SCEVHandle> NewOps(op_begin(), op_end());
2835 NewOps[0] = SE.getNegativeSCEV(SE.getConstant(Range.getUpper()));
2836 SCEVHandle NewAddRec = SE.getAddRecExpr(NewOps, getLoop());
2838 // Next, solve the constructed addrec
2839 std::pair<SCEVHandle,SCEVHandle> Roots =
2840 SolveQuadraticEquation(cast<SCEVAddRecExpr>(NewAddRec), SE);
2841 SCEVConstant *R1 = dyn_cast<SCEVConstant>(Roots.first);
2842 SCEVConstant *R2 = dyn_cast<SCEVConstant>(Roots.second);
2844 // Pick the smallest positive root value.
2845 if (ConstantInt *CB =
2846 dyn_cast<ConstantInt>(ConstantExpr::getICmp(ICmpInst::ICMP_ULT,
2847 R1->getValue(), R2->getValue()))) {
2848 if (CB->getZExtValue() == false)
2849 std::swap(R1, R2); // R1 is the minimum root now.
2851 // Make sure the root is not off by one. The returned iteration should
2852 // not be in the range, but the previous one should be. When solving
2853 // for "X*X < 5", for example, we should not return a root of 2.
2854 ConstantInt *R1Val = EvaluateConstantChrecAtConstant(this,
2857 if (Range.contains(R1Val->getValue())) {
2858 // The next iteration must be out of the range...
2859 ConstantInt *NextVal = ConstantInt::get(R1->getValue()->getValue()+1);
2861 R1Val = EvaluateConstantChrecAtConstant(this, NextVal, SE);
2862 if (!Range.contains(R1Val->getValue()))
2863 return SE.getConstant(NextVal);
2864 return new SCEVCouldNotCompute(); // Something strange happened
2867 // If R1 was not in the range, then it is a good return value. Make
2868 // sure that R1-1 WAS in the range though, just in case.
2869 ConstantInt *NextVal = ConstantInt::get(R1->getValue()->getValue()-1);
2870 R1Val = EvaluateConstantChrecAtConstant(this, NextVal, SE);
2871 if (Range.contains(R1Val->getValue()))
2873 return new SCEVCouldNotCompute(); // Something strange happened
2878 // Fallback, if this is a general polynomial, figure out the progression
2879 // through brute force: evaluate until we find an iteration that fails the
2880 // test. This is likely to be slow, but getting an accurate trip count is
2881 // incredibly important, we will be able to simplify the exit test a lot, and
2882 // we are almost guaranteed to get a trip count in this case.
2883 ConstantInt *TestVal = ConstantInt::get(getType(), 0);
2884 ConstantInt *EndVal = TestVal; // Stop when we wrap around.
2886 ++NumBruteForceEvaluations;
2887 SCEVHandle Val = evaluateAtIteration(SE.getConstant(TestVal), SE);
2888 if (!isa<SCEVConstant>(Val)) // This shouldn't happen.
2889 return new SCEVCouldNotCompute();
2891 // Check to see if we found the value!
2892 if (!Range.contains(cast<SCEVConstant>(Val)->getValue()->getValue()))
2893 return SE.getConstant(TestVal);
2895 // Increment to test the next index.
2896 TestVal = ConstantInt::get(TestVal->getValue()+1);
2897 } while (TestVal != EndVal);
2899 return new SCEVCouldNotCompute();
2904 //===----------------------------------------------------------------------===//
2905 // ScalarEvolution Class Implementation
2906 //===----------------------------------------------------------------------===//
2908 bool ScalarEvolution::runOnFunction(Function &F) {
2909 Impl = new ScalarEvolutionsImpl(*this, F, getAnalysis<LoopInfo>());
2913 void ScalarEvolution::releaseMemory() {
2914 delete (ScalarEvolutionsImpl*)Impl;
2918 void ScalarEvolution::getAnalysisUsage(AnalysisUsage &AU) const {
2919 AU.setPreservesAll();
2920 AU.addRequiredTransitive<LoopInfo>();
2923 SCEVHandle ScalarEvolution::getSCEV(Value *V) const {
2924 return ((ScalarEvolutionsImpl*)Impl)->getSCEV(V);
2927 /// hasSCEV - Return true if the SCEV for this value has already been
2929 bool ScalarEvolution::hasSCEV(Value *V) const {
2930 return ((ScalarEvolutionsImpl*)Impl)->hasSCEV(V);
2934 /// setSCEV - Insert the specified SCEV into the map of current SCEVs for
2935 /// the specified value.
2936 void ScalarEvolution::setSCEV(Value *V, const SCEVHandle &H) {
2937 ((ScalarEvolutionsImpl*)Impl)->setSCEV(V, H);
2941 SCEVHandle ScalarEvolution::getIterationCount(const Loop *L) const {
2942 return ((ScalarEvolutionsImpl*)Impl)->getIterationCount(L);
2945 bool ScalarEvolution::hasLoopInvariantIterationCount(const Loop *L) const {
2946 return !isa<SCEVCouldNotCompute>(getIterationCount(L));
2949 SCEVHandle ScalarEvolution::getSCEVAtScope(Value *V, const Loop *L) const {
2950 return ((ScalarEvolutionsImpl*)Impl)->getSCEVAtScope(getSCEV(V), L);
2953 void ScalarEvolution::deleteValueFromRecords(Value *V) const {
2954 return ((ScalarEvolutionsImpl*)Impl)->deleteValueFromRecords(V);
2957 static void PrintLoopInfo(std::ostream &OS, const ScalarEvolution *SE,
2959 // Print all inner loops first
2960 for (Loop::iterator I = L->begin(), E = L->end(); I != E; ++I)
2961 PrintLoopInfo(OS, SE, *I);
2963 OS << "Loop " << L->getHeader()->getName() << ": ";
2965 SmallVector<BasicBlock*, 8> ExitBlocks;
2966 L->getExitBlocks(ExitBlocks);
2967 if (ExitBlocks.size() != 1)
2968 OS << "<multiple exits> ";
2970 if (SE->hasLoopInvariantIterationCount(L)) {
2971 OS << *SE->getIterationCount(L) << " iterations! ";
2973 OS << "Unpredictable iteration count. ";
2979 void ScalarEvolution::print(std::ostream &OS, const Module* ) const {
2980 Function &F = ((ScalarEvolutionsImpl*)Impl)->F;
2981 LoopInfo &LI = ((ScalarEvolutionsImpl*)Impl)->LI;
2983 OS << "Classifying expressions for: " << F.getName() << "\n";
2984 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
2985 if (I->getType()->isInteger()) {
2988 SCEVHandle SV = getSCEV(&*I);
2992 if (const Loop *L = LI.getLoopFor((*I).getParent())) {
2994 SCEVHandle ExitValue = getSCEVAtScope(&*I, L->getParentLoop());
2995 if (isa<SCEVCouldNotCompute>(ExitValue)) {
2996 OS << "<<Unknown>>";
3006 OS << "Determining loop execution counts for: " << F.getName() << "\n";
3007 for (LoopInfo::iterator I = LI.begin(), E = LI.end(); I != E; ++I)
3008 PrintLoopInfo(OS, this, *I);