blob: d2c3f58e9cfd40e6b5f6f6f0d66e6e426852fdd0 [file] [log] [blame]
//===- ScalarEvolution.cpp - Scalar Evolution Analysis ----------*- C++ -*-===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file contains the implementation of the scalar evolution analysis
// engine, which is used primarily to analyze expressions involving induction
// variables in loops.
//
// There are several aspects to this library. First is the representation of
// scalar expressions, which are represented as subclasses of the SCEV class.
// These classes are used to represent certain types of subexpressions that we
// can handle. We only create one SCEV of a particular shape, so
// pointer-comparisons for equality are legal.
//
// One important aspect of the SCEV objects is that they are never cyclic, even
// if there is a cycle in the dataflow for an expression (ie, a PHI node). If
// the PHI node is one of the idioms that we can represent (e.g., a polynomial
// recurrence) then we represent it directly as a recurrence node, otherwise we
// represent it as a SCEVUnknown node.
//
// In addition to being able to represent expressions of various types, we also
// have folders that are used to build the *canonical* representation for a
// particular expression. These folders are capable of using a variety of
// rewrite rules to simplify the expressions.
//
// Once the folders are defined, we can implement the more interesting
// higher-level code, such as the code that recognizes PHI nodes of various
// types, computes the execution count of a loop, etc.
//
// TODO: We should use these routines and value representations to implement
// dependence analysis!
//
//===----------------------------------------------------------------------===//
//
// There are several good references for the techniques used in this analysis.
//
// Chains of recurrences -- a method to expedite the evaluation
// of closed-form functions
// Olaf Bachmann, Paul S. Wang, Eugene V. Zima
//
// On computational properties of chains of recurrences
// Eugene V. Zima
//
// Symbolic Evaluation of Chains of Recurrences for Loop Optimization
// Robert A. van Engelen
//
// Efficient Symbolic Analysis for Optimizing Compilers
// Robert A. van Engelen
//
// Using the chains of recurrences algebra for data dependence testing and
// induction variable substitution
// MS Thesis, Johnie Birch
//
//===----------------------------------------------------------------------===//
#define DEBUG_TYPE "scalar-evolution"
#include "llvm/Analysis/ScalarEvolutionExpressions.h"
#include "llvm/Constants.h"
#include "llvm/DerivedTypes.h"
#include "llvm/GlobalVariable.h"
#include "llvm/Instructions.h"
#include "llvm/LLVMContext.h"
#include "llvm/Operator.h"
#include "llvm/Analysis/ConstantFolding.h"
#include "llvm/Analysis/Dominators.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/Assembly/Writer.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/ConstantRange.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/GetElementPtrTypeIterator.h"
#include "llvm/Support/InstIterator.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SmallPtrSet.h"
#include <algorithm>
using namespace llvm;
STATISTIC(NumArrayLenItCounts,
"Number of trip counts computed with array length");
STATISTIC(NumTripCountsComputed,
"Number of loops with predictable loop counts");
STATISTIC(NumTripCountsNotComputed,
"Number of loops without predictable loop counts");
STATISTIC(NumBruteForceTripCountsComputed,
"Number of loops with trip counts computed by force");
static cl::opt<unsigned>
MaxBruteForceIterations("scalar-evolution-max-iterations", cl::ReallyHidden,
cl::desc("Maximum number of iterations SCEV will "
"symbolically execute a constant "
"derived loop"),
cl::init(100));
static RegisterPass<ScalarEvolution>
R("scalar-evolution", "Scalar Evolution Analysis", false, true);
char ScalarEvolution::ID = 0;
//===----------------------------------------------------------------------===//
// SCEV class definitions
//===----------------------------------------------------------------------===//
//===----------------------------------------------------------------------===//
// Implementation of the SCEV class.
//
SCEV::~SCEV() {}
void SCEV::dump() const {
print(errs());
errs() << '\n';
}
void SCEV::print(std::ostream &o) const {
raw_os_ostream OS(o);
print(OS);
}
bool SCEV::isZero() const {
if (const SCEVConstant *SC = dyn_cast<SCEVConstant>(this))
return SC->getValue()->isZero();
return false;
}
bool SCEV::isOne() const {
if (const SCEVConstant *SC = dyn_cast<SCEVConstant>(this))
return SC->getValue()->isOne();
return false;
}
bool SCEV::isAllOnesValue() const {
if (const SCEVConstant *SC = dyn_cast<SCEVConstant>(this))
return SC->getValue()->isAllOnesValue();
return false;
}
SCEVCouldNotCompute::SCEVCouldNotCompute() :
SCEV(FoldingSetNodeID(), scCouldNotCompute) {}
bool SCEVCouldNotCompute::isLoopInvariant(const Loop *L) const {
llvm_unreachable("Attempt to use a SCEVCouldNotCompute object!");
return false;
}
const Type *SCEVCouldNotCompute::getType() const {
llvm_unreachable("Attempt to use a SCEVCouldNotCompute object!");
return 0;
}
bool SCEVCouldNotCompute::hasComputableLoopEvolution(const Loop *L) const {
llvm_unreachable("Attempt to use a SCEVCouldNotCompute object!");
return false;
}
bool SCEVCouldNotCompute::hasOperand(const SCEV *) const {
llvm_unreachable("Attempt to use a SCEVCouldNotCompute object!");
return false;
}
void SCEVCouldNotCompute::print(raw_ostream &OS) const {
OS << "***COULDNOTCOMPUTE***";
}
bool SCEVCouldNotCompute::classof(const SCEV *S) {
return S->getSCEVType() == scCouldNotCompute;
}
const SCEV *ScalarEvolution::getConstant(ConstantInt *V) {
FoldingSetNodeID ID;
ID.AddInteger(scConstant);
ID.AddPointer(V);
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVConstant>();
new (S) SCEVConstant(ID, V);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
const SCEV *ScalarEvolution::getConstant(const APInt& Val) {
return getConstant(ConstantInt::get(getContext(), Val));
}
const SCEV *
ScalarEvolution::getConstant(const Type *Ty, uint64_t V, bool isSigned) {
return getConstant(
ConstantInt::get(cast<IntegerType>(Ty), V, isSigned));
}
const Type *SCEVConstant::getType() const { return V->getType(); }
void SCEVConstant::print(raw_ostream &OS) const {
WriteAsOperand(OS, V, false);
}
SCEVCastExpr::SCEVCastExpr(const FoldingSetNodeID &ID,
unsigned SCEVTy, const SCEV *op, const Type *ty)
: SCEV(ID, SCEVTy), Op(op), Ty(ty) {}
bool SCEVCastExpr::dominates(BasicBlock *BB, DominatorTree *DT) const {
return Op->dominates(BB, DT);
}
SCEVTruncateExpr::SCEVTruncateExpr(const FoldingSetNodeID &ID,
const SCEV *op, const Type *ty)
: SCEVCastExpr(ID, scTruncate, op, ty) {
assert((Op->getType()->isInteger() || isa<PointerType>(Op->getType())) &&
(Ty->isInteger() || isa<PointerType>(Ty)) &&
"Cannot truncate non-integer value!");
}
void SCEVTruncateExpr::print(raw_ostream &OS) const {
OS << "(trunc " << *Op->getType() << " " << *Op << " to " << *Ty << ")";
}
SCEVZeroExtendExpr::SCEVZeroExtendExpr(const FoldingSetNodeID &ID,
const SCEV *op, const Type *ty)
: SCEVCastExpr(ID, scZeroExtend, op, ty) {
assert((Op->getType()->isInteger() || isa<PointerType>(Op->getType())) &&
(Ty->isInteger() || isa<PointerType>(Ty)) &&
"Cannot zero extend non-integer value!");
}
void SCEVZeroExtendExpr::print(raw_ostream &OS) const {
OS << "(zext " << *Op->getType() << " " << *Op << " to " << *Ty << ")";
}
SCEVSignExtendExpr::SCEVSignExtendExpr(const FoldingSetNodeID &ID,
const SCEV *op, const Type *ty)
: SCEVCastExpr(ID, scSignExtend, op, ty) {
assert((Op->getType()->isInteger() || isa<PointerType>(Op->getType())) &&
(Ty->isInteger() || isa<PointerType>(Ty)) &&
"Cannot sign extend non-integer value!");
}
void SCEVSignExtendExpr::print(raw_ostream &OS) const {
OS << "(sext " << *Op->getType() << " " << *Op << " to " << *Ty << ")";
}
void SCEVCommutativeExpr::print(raw_ostream &OS) const {
assert(Operands.size() > 1 && "This plus expr shouldn't exist!");
const char *OpStr = getOperationStr();
OS << "(" << *Operands[0];
for (unsigned i = 1, e = Operands.size(); i != e; ++i)
OS << OpStr << *Operands[i];
OS << ")";
}
bool SCEVNAryExpr::dominates(BasicBlock *BB, DominatorTree *DT) const {
for (unsigned i = 0, e = getNumOperands(); i != e; ++i) {
if (!getOperand(i)->dominates(BB, DT))
return false;
}
return true;
}
bool SCEVUDivExpr::dominates(BasicBlock *BB, DominatorTree *DT) const {
return LHS->dominates(BB, DT) && RHS->dominates(BB, DT);
}
void SCEVUDivExpr::print(raw_ostream &OS) const {
OS << "(" << *LHS << " /u " << *RHS << ")";
}
const Type *SCEVUDivExpr::getType() const {
// In most cases the types of LHS and RHS will be the same, but in some
// crazy cases one or the other may be a pointer. ScalarEvolution doesn't
// depend on the type for correctness, but handling types carefully can
// avoid extra casts in the SCEVExpander. The LHS is more likely to be
// a pointer type than the RHS, so use the RHS' type here.
return RHS->getType();
}
bool SCEVAddRecExpr::isLoopInvariant(const Loop *QueryLoop) const {
// Add recurrences are never invariant in the function-body (null loop).
if (!QueryLoop)
return false;
// This recurrence is variant w.r.t. QueryLoop if QueryLoop contains L.
if (QueryLoop->contains(L->getHeader()))
return false;
// This recurrence is variant w.r.t. QueryLoop if any of its operands
// are variant.
for (unsigned i = 0, e = getNumOperands(); i != e; ++i)
if (!getOperand(i)->isLoopInvariant(QueryLoop))
return false;
// Otherwise it's loop-invariant.
return true;
}
void SCEVAddRecExpr::print(raw_ostream &OS) const {
OS << "{" << *Operands[0];
for (unsigned i = 1, e = Operands.size(); i != e; ++i)
OS << ",+," << *Operands[i];
OS << "}<" << L->getHeader()->getName() + ">";
}
void SCEVFieldOffsetExpr::print(raw_ostream &OS) const {
// LLVM struct fields don't have names, so just print the field number.
OS << "offsetof(" << *STy << ", " << FieldNo << ")";
}
void SCEVAllocSizeExpr::print(raw_ostream &OS) const {
OS << "sizeof(" << *AllocTy << ")";
}
bool SCEVUnknown::isLoopInvariant(const Loop *L) const {
// All non-instruction values are loop invariant. All instructions are loop
// invariant if they are not contained in the specified loop.
// Instructions are never considered invariant in the function body
// (null loop) because they are defined within the "loop".
if (Instruction *I = dyn_cast<Instruction>(V))
return L && !L->contains(I->getParent());
return true;
}
bool SCEVUnknown::dominates(BasicBlock *BB, DominatorTree *DT) const {
if (Instruction *I = dyn_cast<Instruction>(getValue()))
return DT->dominates(I->getParent(), BB);
return true;
}
const Type *SCEVUnknown::getType() const {
return V->getType();
}
void SCEVUnknown::print(raw_ostream &OS) const {
WriteAsOperand(OS, V, false);
}
//===----------------------------------------------------------------------===//
// SCEV Utilities
//===----------------------------------------------------------------------===//
static bool CompareTypes(const Type *A, const Type *B) {
if (A->getTypeID() != B->getTypeID())
return A->getTypeID() < B->getTypeID();
if (const IntegerType *AI = dyn_cast<IntegerType>(A)) {
const IntegerType *BI = cast<IntegerType>(B);
return AI->getBitWidth() < BI->getBitWidth();
}
if (const PointerType *AI = dyn_cast<PointerType>(A)) {
const PointerType *BI = cast<PointerType>(B);
return CompareTypes(AI->getElementType(), BI->getElementType());
}
if (const ArrayType *AI = dyn_cast<ArrayType>(A)) {
const ArrayType *BI = cast<ArrayType>(B);
if (AI->getNumElements() != BI->getNumElements())
return AI->getNumElements() < BI->getNumElements();
return CompareTypes(AI->getElementType(), BI->getElementType());
}
if (const VectorType *AI = dyn_cast<VectorType>(A)) {
const VectorType *BI = cast<VectorType>(B);
if (AI->getNumElements() != BI->getNumElements())
return AI->getNumElements() < BI->getNumElements();
return CompareTypes(AI->getElementType(), BI->getElementType());
}
if (const StructType *AI = dyn_cast<StructType>(A)) {
const StructType *BI = cast<StructType>(B);
if (AI->getNumElements() != BI->getNumElements())
return AI->getNumElements() < BI->getNumElements();
for (unsigned i = 0, e = AI->getNumElements(); i != e; ++i)
if (CompareTypes(AI->getElementType(i), BI->getElementType(i)) ||
CompareTypes(BI->getElementType(i), AI->getElementType(i)))
return CompareTypes(AI->getElementType(i), BI->getElementType(i));
}
return false;
}
namespace {
/// SCEVComplexityCompare - Return true if the complexity of the LHS is less
/// than the complexity of the RHS. This comparator is used to canonicalize
/// expressions.
class VISIBILITY_HIDDEN SCEVComplexityCompare {
LoopInfo *LI;
public:
explicit SCEVComplexityCompare(LoopInfo *li) : LI(li) {}
bool operator()(const SCEV *LHS, const SCEV *RHS) const {
// Primarily, sort the SCEVs by their getSCEVType().
if (LHS->getSCEVType() != RHS->getSCEVType())
return LHS->getSCEVType() < RHS->getSCEVType();
// Aside from the getSCEVType() ordering, the particular ordering
// isn't very important except that it's beneficial to be consistent,
// so that (a + b) and (b + a) don't end up as different expressions.
// Sort SCEVUnknown values with some loose heuristics. TODO: This is
// not as complete as it could be.
if (const SCEVUnknown *LU = dyn_cast<SCEVUnknown>(LHS)) {
const SCEVUnknown *RU = cast<SCEVUnknown>(RHS);
// Order pointer values after integer values. This helps SCEVExpander
// form GEPs.
if (isa<PointerType>(LU->getType()) && !isa<PointerType>(RU->getType()))
return false;
if (isa<PointerType>(RU->getType()) && !isa<PointerType>(LU->getType()))
return true;
// Compare getValueID values.
if (LU->getValue()->getValueID() != RU->getValue()->getValueID())
return LU->getValue()->getValueID() < RU->getValue()->getValueID();
// Sort arguments by their position.
if (const Argument *LA = dyn_cast<Argument>(LU->getValue())) {
const Argument *RA = cast<Argument>(RU->getValue());
return LA->getArgNo() < RA->getArgNo();
}
// For instructions, compare their loop depth, and their opcode.
// This is pretty loose.
if (Instruction *LV = dyn_cast<Instruction>(LU->getValue())) {
Instruction *RV = cast<Instruction>(RU->getValue());
// Compare loop depths.
if (LI->getLoopDepth(LV->getParent()) !=
LI->getLoopDepth(RV->getParent()))
return LI->getLoopDepth(LV->getParent()) <
LI->getLoopDepth(RV->getParent());
// Compare opcodes.
if (LV->getOpcode() != RV->getOpcode())
return LV->getOpcode() < RV->getOpcode();
// Compare the number of operands.
if (LV->getNumOperands() != RV->getNumOperands())
return LV->getNumOperands() < RV->getNumOperands();
}
return false;
}
// Compare constant values.
if (const SCEVConstant *LC = dyn_cast<SCEVConstant>(LHS)) {
const SCEVConstant *RC = cast<SCEVConstant>(RHS);
if (LC->getValue()->getBitWidth() != RC->getValue()->getBitWidth())
return LC->getValue()->getBitWidth() < RC->getValue()->getBitWidth();
return LC->getValue()->getValue().ult(RC->getValue()->getValue());
}
// Compare addrec loop depths.
if (const SCEVAddRecExpr *LA = dyn_cast<SCEVAddRecExpr>(LHS)) {
const SCEVAddRecExpr *RA = cast<SCEVAddRecExpr>(RHS);
if (LA->getLoop()->getLoopDepth() != RA->getLoop()->getLoopDepth())
return LA->getLoop()->getLoopDepth() < RA->getLoop()->getLoopDepth();
}
// Lexicographically compare n-ary expressions.
if (const SCEVNAryExpr *LC = dyn_cast<SCEVNAryExpr>(LHS)) {
const SCEVNAryExpr *RC = cast<SCEVNAryExpr>(RHS);
for (unsigned i = 0, e = LC->getNumOperands(); i != e; ++i) {
if (i >= RC->getNumOperands())
return false;
if (operator()(LC->getOperand(i), RC->getOperand(i)))
return true;
if (operator()(RC->getOperand(i), LC->getOperand(i)))
return false;
}
return LC->getNumOperands() < RC->getNumOperands();
}
// Lexicographically compare udiv expressions.
if (const SCEVUDivExpr *LC = dyn_cast<SCEVUDivExpr>(LHS)) {
const SCEVUDivExpr *RC = cast<SCEVUDivExpr>(RHS);
if (operator()(LC->getLHS(), RC->getLHS()))
return true;
if (operator()(RC->getLHS(), LC->getLHS()))
return false;
if (operator()(LC->getRHS(), RC->getRHS()))
return true;
if (operator()(RC->getRHS(), LC->getRHS()))
return false;
return false;
}
// Compare cast expressions by operand.
if (const SCEVCastExpr *LC = dyn_cast<SCEVCastExpr>(LHS)) {
const SCEVCastExpr *RC = cast<SCEVCastExpr>(RHS);
return operator()(LC->getOperand(), RC->getOperand());
}
// Compare offsetof expressions.
if (const SCEVFieldOffsetExpr *LA = dyn_cast<SCEVFieldOffsetExpr>(LHS)) {
const SCEVFieldOffsetExpr *RA = cast<SCEVFieldOffsetExpr>(RHS);
if (CompareTypes(LA->getStructType(), RA->getStructType()) ||
CompareTypes(RA->getStructType(), LA->getStructType()))
return CompareTypes(LA->getStructType(), RA->getStructType());
return LA->getFieldNo() < RA->getFieldNo();
}
// Compare sizeof expressions by the allocation type.
if (const SCEVAllocSizeExpr *LA = dyn_cast<SCEVAllocSizeExpr>(LHS)) {
const SCEVAllocSizeExpr *RA = cast<SCEVAllocSizeExpr>(RHS);
return CompareTypes(LA->getAllocType(), RA->getAllocType());
}
llvm_unreachable("Unknown SCEV kind!");
return false;
}
};
}
/// GroupByComplexity - Given a list of SCEV objects, order them by their
/// complexity, and group objects of the same complexity together by value.
/// When this routine is finished, we know that any duplicates in the vector are
/// consecutive and that complexity is monotonically increasing.
///
/// Note that we go take special precautions to ensure that we get determinstic
/// results from this routine. In other words, we don't want the results of
/// this to depend on where the addresses of various SCEV objects happened to
/// land in memory.
///
static void GroupByComplexity(SmallVectorImpl<const SCEV *> &Ops,
LoopInfo *LI) {
if (Ops.size() < 2) return; // Noop
if (Ops.size() == 2) {
// This is the common case, which also happens to be trivially simple.
// Special case it.
if (SCEVComplexityCompare(LI)(Ops[1], Ops[0]))
std::swap(Ops[0], Ops[1]);
return;
}
// Do the rough sort by complexity.
std::stable_sort(Ops.begin(), Ops.end(), SCEVComplexityCompare(LI));
// Now that we are sorted by complexity, group elements of the same
// complexity. Note that this is, at worst, N^2, but the vector is likely to
// be extremely short in practice. Note that we take this approach because we
// do not want to depend on the addresses of the objects we are grouping.
for (unsigned i = 0, e = Ops.size(); i != e-2; ++i) {
const SCEV *S = Ops[i];
unsigned Complexity = S->getSCEVType();
// If there are any objects of the same complexity and same value as this
// one, group them.
for (unsigned j = i+1; j != e && Ops[j]->getSCEVType() == Complexity; ++j) {
if (Ops[j] == S) { // Found a duplicate.
// Move it to immediately after i'th element.
std::swap(Ops[i+1], Ops[j]);
++i; // no need to rescan it.
if (i == e-2) return; // Done!
}
}
}
}
//===----------------------------------------------------------------------===//
// Simple SCEV method implementations
//===----------------------------------------------------------------------===//
/// BinomialCoefficient - Compute BC(It, K). The result has width W.
/// Assume, K > 0.
static const SCEV *BinomialCoefficient(const SCEV *It, unsigned K,
ScalarEvolution &SE,
const Type* ResultTy) {
// Handle the simplest case efficiently.
if (K == 1)
return SE.getTruncateOrZeroExtend(It, ResultTy);
// We are using the following formula for BC(It, K):
//
// BC(It, K) = (It * (It - 1) * ... * (It - K + 1)) / K!
//
// Suppose, W is the bitwidth of the return value. We must be prepared for
// overflow. Hence, we must assure that the result of our computation is
// equal to the accurate one modulo 2^W. Unfortunately, division isn't
// safe in modular arithmetic.
//
// However, this code doesn't use exactly that formula; the formula it uses
// is something like the following, where T is the number of factors of 2 in
// K! (i.e. trailing zeros in the binary representation of K!), and ^ is
// exponentiation:
//
// BC(It, K) = (It * (It - 1) * ... * (It - K + 1)) / 2^T / (K! / 2^T)
//
// This formula is trivially equivalent to the previous formula. However,
// this formula can be implemented much more efficiently. The trick is that
// K! / 2^T is odd, and exact division by an odd number *is* safe in modular
// arithmetic. To do exact division in modular arithmetic, all we have
// to do is multiply by the inverse. Therefore, this step can be done at
// width W.
//
// The next issue is how to safely do the division by 2^T. The way this
// is done is by doing the multiplication step at a width of at least W + T
// bits. This way, the bottom W+T bits of the product are accurate. Then,
// when we perform the division by 2^T (which is equivalent to a right shift
// by T), the bottom W bits are accurate. Extra bits are okay; they'll get
// truncated out after the division by 2^T.
//
// In comparison to just directly using the first formula, this technique
// is much more efficient; using the first formula requires W * K bits,
// but this formula less than W + K bits. Also, the first formula requires
// a division step, whereas this formula only requires multiplies and shifts.
//
// It doesn't matter whether the subtraction step is done in the calculation
// width or the input iteration count's width; if the subtraction overflows,
// the result must be zero anyway. We prefer here to do it in the width of
// the induction variable because it helps a lot for certain cases; CodeGen
// isn't smart enough to ignore the overflow, which leads to much less
// efficient code if the width of the subtraction is wider than the native
// register width.
//
// (It's possible to not widen at all by pulling out factors of 2 before
// the multiplication; for example, K=2 can be calculated as
// It/2*(It+(It*INT_MIN/INT_MIN)+-1). However, it requires
// extra arithmetic, so it's not an obvious win, and it gets
// much more complicated for K > 3.)
// Protection from insane SCEVs; this bound is conservative,
// but it probably doesn't matter.
if (K > 1000)
return SE.getCouldNotCompute();
unsigned W = SE.getTypeSizeInBits(ResultTy);
// Calculate K! / 2^T and T; we divide out the factors of two before
// multiplying for calculating K! / 2^T to avoid overflow.
// Other overflow doesn't matter because we only care about the bottom
// W bits of the result.
APInt OddFactorial(W, 1);
unsigned T = 1;
for (unsigned i = 3; i <= K; ++i) {
APInt Mult(W, i);
unsigned TwoFactors = Mult.countTrailingZeros();
T += TwoFactors;
Mult = Mult.lshr(TwoFactors);
OddFactorial *= Mult;
}
// We need at least W + T bits for the multiplication step
unsigned CalculationBits = W + T;
// Calcuate 2^T, at width T+W.
APInt DivFactor = APInt(CalculationBits, 1).shl(T);
// Calculate the multiplicative inverse of K! / 2^T;
// this multiplication factor will perform the exact division by
// K! / 2^T.
APInt Mod = APInt::getSignedMinValue(W+1);
APInt MultiplyFactor = OddFactorial.zext(W+1);
MultiplyFactor = MultiplyFactor.multiplicativeInverse(Mod);
MultiplyFactor = MultiplyFactor.trunc(W);
// Calculate the product, at width T+W
const IntegerType *CalculationTy = IntegerType::get(SE.getContext(),
CalculationBits);
const SCEV *Dividend = SE.getTruncateOrZeroExtend(It, CalculationTy);
for (unsigned i = 1; i != K; ++i) {
const SCEV *S = SE.getMinusSCEV(It, SE.getIntegerSCEV(i, It->getType()));
Dividend = SE.getMulExpr(Dividend,
SE.getTruncateOrZeroExtend(S, CalculationTy));
}
// Divide by 2^T
const SCEV *DivResult = SE.getUDivExpr(Dividend, SE.getConstant(DivFactor));
// Truncate the result, and divide by K! / 2^T.
return SE.getMulExpr(SE.getConstant(MultiplyFactor),
SE.getTruncateOrZeroExtend(DivResult, ResultTy));
}
/// evaluateAtIteration - Return the value of this chain of recurrences at
/// the specified iteration number. We can evaluate this recurrence by
/// multiplying each element in the chain by the binomial coefficient
/// corresponding to it. In other words, we can evaluate {A,+,B,+,C,+,D} as:
///
/// A*BC(It, 0) + B*BC(It, 1) + C*BC(It, 2) + D*BC(It, 3)
///
/// where BC(It, k) stands for binomial coefficient.
///
const SCEV *SCEVAddRecExpr::evaluateAtIteration(const SCEV *It,
ScalarEvolution &SE) const {
const SCEV *Result = getStart();
for (unsigned i = 1, e = getNumOperands(); i != e; ++i) {
// The computation is correct in the face of overflow provided that the
// multiplication is performed _after_ the evaluation of the binomial
// coefficient.
const SCEV *Coeff = BinomialCoefficient(It, i, SE, getType());
if (isa<SCEVCouldNotCompute>(Coeff))
return Coeff;
Result = SE.getAddExpr(Result, SE.getMulExpr(getOperand(i), Coeff));
}
return Result;
}
//===----------------------------------------------------------------------===//
// SCEV Expression folder implementations
//===----------------------------------------------------------------------===//
const SCEV *ScalarEvolution::getTruncateExpr(const SCEV *Op,
const Type *Ty) {
assert(getTypeSizeInBits(Op->getType()) > getTypeSizeInBits(Ty) &&
"This is not a truncating conversion!");
assert(isSCEVable(Ty) &&
"This is not a conversion to a SCEVable type!");
Ty = getEffectiveSCEVType(Ty);
FoldingSetNodeID ID;
ID.AddInteger(scTruncate);
ID.AddPointer(Op);
ID.AddPointer(Ty);
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
// Fold if the operand is constant.
if (const SCEVConstant *SC = dyn_cast<SCEVConstant>(Op))
return getConstant(
cast<ConstantInt>(ConstantExpr::getTrunc(SC->getValue(), Ty)));
// trunc(trunc(x)) --> trunc(x)
if (const SCEVTruncateExpr *ST = dyn_cast<SCEVTruncateExpr>(Op))
return getTruncateExpr(ST->getOperand(), Ty);
// trunc(sext(x)) --> sext(x) if widening or trunc(x) if narrowing
if (const SCEVSignExtendExpr *SS = dyn_cast<SCEVSignExtendExpr>(Op))
return getTruncateOrSignExtend(SS->getOperand(), Ty);
// trunc(zext(x)) --> zext(x) if widening or trunc(x) if narrowing
if (const SCEVZeroExtendExpr *SZ = dyn_cast<SCEVZeroExtendExpr>(Op))
return getTruncateOrZeroExtend(SZ->getOperand(), Ty);
// If the input value is a chrec scev, truncate the chrec's operands.
if (const SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(Op)) {
SmallVector<const SCEV *, 4> Operands;
for (unsigned i = 0, e = AddRec->getNumOperands(); i != e; ++i)
Operands.push_back(getTruncateExpr(AddRec->getOperand(i), Ty));
return getAddRecExpr(Operands, AddRec->getLoop());
}
// The cast wasn't folded; create an explicit cast node.
// Recompute the insert position, as it may have been invalidated.
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVTruncateExpr>();
new (S) SCEVTruncateExpr(ID, Op, Ty);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
const SCEV *ScalarEvolution::getZeroExtendExpr(const SCEV *Op,
const Type *Ty) {
assert(getTypeSizeInBits(Op->getType()) < getTypeSizeInBits(Ty) &&
"This is not an extending conversion!");
assert(isSCEVable(Ty) &&
"This is not a conversion to a SCEVable type!");
Ty = getEffectiveSCEVType(Ty);
// Fold if the operand is constant.
if (const SCEVConstant *SC = dyn_cast<SCEVConstant>(Op)) {
const Type *IntTy = getEffectiveSCEVType(Ty);
Constant *C = ConstantExpr::getZExt(SC->getValue(), IntTy);
if (IntTy != Ty) C = ConstantExpr::getIntToPtr(C, Ty);
return getConstant(cast<ConstantInt>(C));
}
// zext(zext(x)) --> zext(x)
if (const SCEVZeroExtendExpr *SZ = dyn_cast<SCEVZeroExtendExpr>(Op))
return getZeroExtendExpr(SZ->getOperand(), Ty);
// Before doing any expensive analysis, check to see if we've already
// computed a SCEV for this Op and Ty.
FoldingSetNodeID ID;
ID.AddInteger(scZeroExtend);
ID.AddPointer(Op);
ID.AddPointer(Ty);
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
// If the input value is a chrec scev, and we can prove that the value
// did not overflow the old, smaller, value, we can zero extend all of the
// operands (often constants). This allows analysis of something like
// this: for (unsigned char X = 0; X < 100; ++X) { int Y = X; }
if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(Op))
if (AR->isAffine()) {
const SCEV *Start = AR->getStart();
const SCEV *Step = AR->getStepRecurrence(*this);
unsigned BitWidth = getTypeSizeInBits(AR->getType());
const Loop *L = AR->getLoop();
// If we have special knowledge that this addrec won't overflow,
// we don't need to do any further analysis.
if (AR->hasNoUnsignedWrap())
return getAddRecExpr(getZeroExtendExpr(Start, Ty),
getZeroExtendExpr(Step, Ty),
L);
// Check whether the backedge-taken count is SCEVCouldNotCompute.
// Note that this serves two purposes: It filters out loops that are
// simply not analyzable, and it covers the case where this code is
// being called from within backedge-taken count analysis, such that
// attempting to ask for the backedge-taken count would likely result
// in infinite recursion. In the later case, the analysis code will
// cope with a conservative value, and it will take care to purge
// that value once it has finished.
const SCEV *MaxBECount = getMaxBackedgeTakenCount(L);
if (!isa<SCEVCouldNotCompute>(MaxBECount)) {
// Manually compute the final value for AR, checking for
// overflow.
// Check whether the backedge-taken count can be losslessly casted to
// the addrec's type. The count is always unsigned.
const SCEV *CastedMaxBECount =
getTruncateOrZeroExtend(MaxBECount, Start->getType());
const SCEV *RecastedMaxBECount =
getTruncateOrZeroExtend(CastedMaxBECount, MaxBECount->getType());
if (MaxBECount == RecastedMaxBECount) {
const Type *WideTy = IntegerType::get(getContext(), BitWidth * 2);
// Check whether Start+Step*MaxBECount has no unsigned overflow.
const SCEV *ZMul =
getMulExpr(CastedMaxBECount,
getTruncateOrZeroExtend(Step, Start->getType()));
const SCEV *Add = getAddExpr(Start, ZMul);
const SCEV *OperandExtendedAdd =
getAddExpr(getZeroExtendExpr(Start, WideTy),
getMulExpr(getZeroExtendExpr(CastedMaxBECount, WideTy),
getZeroExtendExpr(Step, WideTy)));
if (getZeroExtendExpr(Add, WideTy) == OperandExtendedAdd)
// Return the expression with the addrec on the outside.
return getAddRecExpr(getZeroExtendExpr(Start, Ty),
getZeroExtendExpr(Step, Ty),
L);
// Similar to above, only this time treat the step value as signed.
// This covers loops that count down.
const SCEV *SMul =
getMulExpr(CastedMaxBECount,
getTruncateOrSignExtend(Step, Start->getType()));
Add = getAddExpr(Start, SMul);
OperandExtendedAdd =
getAddExpr(getZeroExtendExpr(Start, WideTy),
getMulExpr(getZeroExtendExpr(CastedMaxBECount, WideTy),
getSignExtendExpr(Step, WideTy)));
if (getZeroExtendExpr(Add, WideTy) == OperandExtendedAdd)
// Return the expression with the addrec on the outside.
return getAddRecExpr(getZeroExtendExpr(Start, Ty),
getSignExtendExpr(Step, Ty),
L);
}
// If the backedge is guarded by a comparison with the pre-inc value
// the addrec is safe. Also, if the entry is guarded by a comparison
// with the start value and the backedge is guarded by a comparison
// with the post-inc value, the addrec is safe.
if (isKnownPositive(Step)) {
const SCEV *N = getConstant(APInt::getMinValue(BitWidth) -
getUnsignedRange(Step).getUnsignedMax());
if (isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_ULT, AR, N) ||
(isLoopGuardedByCond(L, ICmpInst::ICMP_ULT, Start, N) &&
isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_ULT,
AR->getPostIncExpr(*this), N)))
// Return the expression with the addrec on the outside.
return getAddRecExpr(getZeroExtendExpr(Start, Ty),
getZeroExtendExpr(Step, Ty),
L);
} else if (isKnownNegative(Step)) {
const SCEV *N = getConstant(APInt::getMaxValue(BitWidth) -
getSignedRange(Step).getSignedMin());
if (isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_UGT, AR, N) &&
(isLoopGuardedByCond(L, ICmpInst::ICMP_UGT, Start, N) ||
isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_UGT,
AR->getPostIncExpr(*this), N)))
// Return the expression with the addrec on the outside.
return getAddRecExpr(getZeroExtendExpr(Start, Ty),
getSignExtendExpr(Step, Ty),
L);
}
}
}
// The cast wasn't folded; create an explicit cast node.
// Recompute the insert position, as it may have been invalidated.
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVZeroExtendExpr>();
new (S) SCEVZeroExtendExpr(ID, Op, Ty);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
const SCEV *ScalarEvolution::getSignExtendExpr(const SCEV *Op,
const Type *Ty) {
assert(getTypeSizeInBits(Op->getType()) < getTypeSizeInBits(Ty) &&
"This is not an extending conversion!");
assert(isSCEVable(Ty) &&
"This is not a conversion to a SCEVable type!");
Ty = getEffectiveSCEVType(Ty);
// Fold if the operand is constant.
if (const SCEVConstant *SC = dyn_cast<SCEVConstant>(Op)) {
const Type *IntTy = getEffectiveSCEVType(Ty);
Constant *C = ConstantExpr::getSExt(SC->getValue(), IntTy);
if (IntTy != Ty) C = ConstantExpr::getIntToPtr(C, Ty);
return getConstant(cast<ConstantInt>(C));
}
// sext(sext(x)) --> sext(x)
if (const SCEVSignExtendExpr *SS = dyn_cast<SCEVSignExtendExpr>(Op))
return getSignExtendExpr(SS->getOperand(), Ty);
// Before doing any expensive analysis, check to see if we've already
// computed a SCEV for this Op and Ty.
FoldingSetNodeID ID;
ID.AddInteger(scSignExtend);
ID.AddPointer(Op);
ID.AddPointer(Ty);
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
// If the input value is a chrec scev, and we can prove that the value
// did not overflow the old, smaller, value, we can sign extend all of the
// operands (often constants). This allows analysis of something like
// this: for (signed char X = 0; X < 100; ++X) { int Y = X; }
if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(Op))
if (AR->isAffine()) {
const SCEV *Start = AR->getStart();
const SCEV *Step = AR->getStepRecurrence(*this);
unsigned BitWidth = getTypeSizeInBits(AR->getType());
const Loop *L = AR->getLoop();
// If we have special knowledge that this addrec won't overflow,
// we don't need to do any further analysis.
if (AR->hasNoSignedWrap())
return getAddRecExpr(getSignExtendExpr(Start, Ty),
getSignExtendExpr(Step, Ty),
L);
// Check whether the backedge-taken count is SCEVCouldNotCompute.
// Note that this serves two purposes: It filters out loops that are
// simply not analyzable, and it covers the case where this code is
// being called from within backedge-taken count analysis, such that
// attempting to ask for the backedge-taken count would likely result
// in infinite recursion. In the later case, the analysis code will
// cope with a conservative value, and it will take care to purge
// that value once it has finished.
const SCEV *MaxBECount = getMaxBackedgeTakenCount(L);
if (!isa<SCEVCouldNotCompute>(MaxBECount)) {
// Manually compute the final value for AR, checking for
// overflow.
// Check whether the backedge-taken count can be losslessly casted to
// the addrec's type. The count is always unsigned.
const SCEV *CastedMaxBECount =
getTruncateOrZeroExtend(MaxBECount, Start->getType());
const SCEV *RecastedMaxBECount =
getTruncateOrZeroExtend(CastedMaxBECount, MaxBECount->getType());
if (MaxBECount == RecastedMaxBECount) {
const Type *WideTy = IntegerType::get(getContext(), BitWidth * 2);
// Check whether Start+Step*MaxBECount has no signed overflow.
const SCEV *SMul =
getMulExpr(CastedMaxBECount,
getTruncateOrSignExtend(Step, Start->getType()));
const SCEV *Add = getAddExpr(Start, SMul);
const SCEV *OperandExtendedAdd =
getAddExpr(getSignExtendExpr(Start, WideTy),
getMulExpr(getZeroExtendExpr(CastedMaxBECount, WideTy),
getSignExtendExpr(Step, WideTy)));
if (getSignExtendExpr(Add, WideTy) == OperandExtendedAdd)
// Return the expression with the addrec on the outside.
return getAddRecExpr(getSignExtendExpr(Start, Ty),
getSignExtendExpr(Step, Ty),
L);
// Similar to above, only this time treat the step value as unsigned.
// This covers loops that count up with an unsigned step.
const SCEV *UMul =
getMulExpr(CastedMaxBECount,
getTruncateOrZeroExtend(Step, Start->getType()));
Add = getAddExpr(Start, UMul);
OperandExtendedAdd =
getAddExpr(getSignExtendExpr(Start, WideTy),
getMulExpr(getZeroExtendExpr(CastedMaxBECount, WideTy),
getZeroExtendExpr(Step, WideTy)));
if (getSignExtendExpr(Add, WideTy) == OperandExtendedAdd)
// Return the expression with the addrec on the outside.
return getAddRecExpr(getSignExtendExpr(Start, Ty),
getZeroExtendExpr(Step, Ty),
L);
}
// If the backedge is guarded by a comparison with the pre-inc value
// the addrec is safe. Also, if the entry is guarded by a comparison
// with the start value and the backedge is guarded by a comparison
// with the post-inc value, the addrec is safe.
if (isKnownPositive(Step)) {
const SCEV *N = getConstant(APInt::getSignedMinValue(BitWidth) -
getSignedRange(Step).getSignedMax());
if (isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_SLT, AR, N) ||
(isLoopGuardedByCond(L, ICmpInst::ICMP_SLT, Start, N) &&
isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_SLT,
AR->getPostIncExpr(*this), N)))
// Return the expression with the addrec on the outside.
return getAddRecExpr(getSignExtendExpr(Start, Ty),
getSignExtendExpr(Step, Ty),
L);
} else if (isKnownNegative(Step)) {
const SCEV *N = getConstant(APInt::getSignedMaxValue(BitWidth) -
getSignedRange(Step).getSignedMin());
if (isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_SGT, AR, N) ||
(isLoopGuardedByCond(L, ICmpInst::ICMP_SGT, Start, N) &&
isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_SGT,
AR->getPostIncExpr(*this), N)))
// Return the expression with the addrec on the outside.
return getAddRecExpr(getSignExtendExpr(Start, Ty),
getSignExtendExpr(Step, Ty),
L);
}
}
}
// The cast wasn't folded; create an explicit cast node.
// Recompute the insert position, as it may have been invalidated.
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVSignExtendExpr>();
new (S) SCEVSignExtendExpr(ID, Op, Ty);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
/// getAnyExtendExpr - Return a SCEV for the given operand extended with
/// unspecified bits out to the given type.
///
const SCEV *ScalarEvolution::getAnyExtendExpr(const SCEV *Op,
const Type *Ty) {
assert(getTypeSizeInBits(Op->getType()) < getTypeSizeInBits(Ty) &&
"This is not an extending conversion!");
assert(isSCEVable(Ty) &&
"This is not a conversion to a SCEVable type!");
Ty = getEffectiveSCEVType(Ty);
// Sign-extend negative constants.
if (const SCEVConstant *SC = dyn_cast<SCEVConstant>(Op))
if (SC->getValue()->getValue().isNegative())
return getSignExtendExpr(Op, Ty);
// Peel off a truncate cast.
if (const SCEVTruncateExpr *T = dyn_cast<SCEVTruncateExpr>(Op)) {
const SCEV *NewOp = T->getOperand();
if (getTypeSizeInBits(NewOp->getType()) < getTypeSizeInBits(Ty))
return getAnyExtendExpr(NewOp, Ty);
return getTruncateOrNoop(NewOp, Ty);
}
// Next try a zext cast. If the cast is folded, use it.
const SCEV *ZExt = getZeroExtendExpr(Op, Ty);
if (!isa<SCEVZeroExtendExpr>(ZExt))
return ZExt;
// Next try a sext cast. If the cast is folded, use it.
const SCEV *SExt = getSignExtendExpr(Op, Ty);
if (!isa<SCEVSignExtendExpr>(SExt))
return SExt;
// If the expression is obviously signed, use the sext cast value.
if (isa<SCEVSMaxExpr>(Op))
return SExt;
// Absent any other information, use the zext cast value.
return ZExt;
}
/// CollectAddOperandsWithScales - Process the given Ops list, which is
/// a list of operands to be added under the given scale, update the given
/// map. This is a helper function for getAddRecExpr. As an example of
/// what it does, given a sequence of operands that would form an add
/// expression like this:
///
/// m + n + 13 + (A * (o + p + (B * q + m + 29))) + r + (-1 * r)
///
/// where A and B are constants, update the map with these values:
///
/// (m, 1+A*B), (n, 1), (o, A), (p, A), (q, A*B), (r, 0)
///
/// and add 13 + A*B*29 to AccumulatedConstant.
/// This will allow getAddRecExpr to produce this:
///
/// 13+A*B*29 + n + (m * (1+A*B)) + ((o + p) * A) + (q * A*B)
///
/// This form often exposes folding opportunities that are hidden in
/// the original operand list.
///
/// Return true iff it appears that any interesting folding opportunities
/// may be exposed. This helps getAddRecExpr short-circuit extra work in
/// the common case where no interesting opportunities are present, and
/// is also used as a check to avoid infinite recursion.
///
static bool
CollectAddOperandsWithScales(DenseMap<const SCEV *, APInt> &M,
SmallVector<const SCEV *, 8> &NewOps,
APInt &AccumulatedConstant,
const SmallVectorImpl<const SCEV *> &Ops,
const APInt &Scale,
ScalarEvolution &SE) {
bool Interesting = false;
// Iterate over the add operands.
for (unsigned i = 0, e = Ops.size(); i != e; ++i) {
const SCEVMulExpr *Mul = dyn_cast<SCEVMulExpr>(Ops[i]);
if (Mul && isa<SCEVConstant>(Mul->getOperand(0))) {
APInt NewScale =
Scale * cast<SCEVConstant>(Mul->getOperand(0))->getValue()->getValue();
if (Mul->getNumOperands() == 2 && isa<SCEVAddExpr>(Mul->getOperand(1))) {
// A multiplication of a constant with another add; recurse.
Interesting |=
CollectAddOperandsWithScales(M, NewOps, AccumulatedConstant,
cast<SCEVAddExpr>(Mul->getOperand(1))
->getOperands(),
NewScale, SE);
} else {
// A multiplication of a constant with some other value. Update
// the map.
SmallVector<const SCEV *, 4> MulOps(Mul->op_begin()+1, Mul->op_end());
const SCEV *Key = SE.getMulExpr(MulOps);
std::pair<DenseMap<const SCEV *, APInt>::iterator, bool> Pair =
M.insert(std::make_pair(Key, NewScale));
if (Pair.second) {
NewOps.push_back(Pair.first->first);
} else {
Pair.first->second += NewScale;
// The map already had an entry for this value, which may indicate
// a folding opportunity.
Interesting = true;
}
}
} else if (const SCEVConstant *C = dyn_cast<SCEVConstant>(Ops[i])) {
// Pull a buried constant out to the outside.
if (Scale != 1 || AccumulatedConstant != 0 || C->isZero())
Interesting = true;
AccumulatedConstant += Scale * C->getValue()->getValue();
} else {
// An ordinary operand. Update the map.
std::pair<DenseMap<const SCEV *, APInt>::iterator, bool> Pair =
M.insert(std::make_pair(Ops[i], Scale));
if (Pair.second) {
NewOps.push_back(Pair.first->first);
} else {
Pair.first->second += Scale;
// The map already had an entry for this value, which may indicate
// a folding opportunity.
Interesting = true;
}
}
}
return Interesting;
}
namespace {
struct APIntCompare {
bool operator()(const APInt &LHS, const APInt &RHS) const {
return LHS.ult(RHS);
}
};
}
/// getAddExpr - Get a canonical add expression, or something simpler if
/// possible.
const SCEV *ScalarEvolution::getAddExpr(SmallVectorImpl<const SCEV *> &Ops) {
assert(!Ops.empty() && "Cannot get empty add!");
if (Ops.size() == 1) return Ops[0];
#ifndef NDEBUG
for (unsigned i = 1, e = Ops.size(); i != e; ++i)
assert(getEffectiveSCEVType(Ops[i]->getType()) ==
getEffectiveSCEVType(Ops[0]->getType()) &&
"SCEVAddExpr operand types don't match!");
#endif
// Sort by complexity, this groups all similar expression types together.
GroupByComplexity(Ops, LI);
// If there are any constants, fold them together.
unsigned Idx = 0;
if (const SCEVConstant *LHSC = dyn_cast<SCEVConstant>(Ops[0])) {
++Idx;
assert(Idx < Ops.size());
while (const SCEVConstant *RHSC = dyn_cast<SCEVConstant>(Ops[Idx])) {
// We found two constants, fold them together!
Ops[0] = getConstant(LHSC->getValue()->getValue() +
RHSC->getValue()->getValue());
if (Ops.size() == 2) return Ops[0];
Ops.erase(Ops.begin()+1); // Erase the folded element
LHSC = cast<SCEVConstant>(Ops[0]);
}
// If we are left with a constant zero being added, strip it off.
if (cast<SCEVConstant>(Ops[0])->getValue()->isZero()) {
Ops.erase(Ops.begin());
--Idx;
}
}
if (Ops.size() == 1) return Ops[0];
// Okay, check to see if the same value occurs in the operand list twice. If
// so, merge them together into an multiply expression. Since we sorted the
// list, these values are required to be adjacent.
const Type *Ty = Ops[0]->getType();
for (unsigned i = 0, e = Ops.size()-1; i != e; ++i)
if (Ops[i] == Ops[i+1]) { // X + Y + Y --> X + Y*2
// Found a match, merge the two values into a multiply, and add any
// remaining values to the result.
const SCEV *Two = getIntegerSCEV(2, Ty);
const SCEV *Mul = getMulExpr(Ops[i], Two);
if (Ops.size() == 2)
return Mul;
Ops.erase(Ops.begin()+i, Ops.begin()+i+2);
Ops.push_back(Mul);
return getAddExpr(Ops);
}
// Check for truncates. If all the operands are truncated from the same
// type, see if factoring out the truncate would permit the result to be
// folded. eg., trunc(x) + m*trunc(n) --> trunc(x + trunc(m)*n)
// if the contents of the resulting outer trunc fold to something simple.
for (; Idx < Ops.size() && isa<SCEVTruncateExpr>(Ops[Idx]); ++Idx) {
const SCEVTruncateExpr *Trunc = cast<SCEVTruncateExpr>(Ops[Idx]);
const Type *DstType = Trunc->getType();
const Type *SrcType = Trunc->getOperand()->getType();
SmallVector<const SCEV *, 8> LargeOps;
bool Ok = true;
// Check all the operands to see if they can be represented in the
// source type of the truncate.
for (unsigned i = 0, e = Ops.size(); i != e; ++i) {
if (const SCEVTruncateExpr *T = dyn_cast<SCEVTruncateExpr>(Ops[i])) {
if (T->getOperand()->getType() != SrcType) {
Ok = false;
break;
}
LargeOps.push_back(T->getOperand());
} else if (const SCEVConstant *C = dyn_cast<SCEVConstant>(Ops[i])) {
// This could be either sign or zero extension, but sign extension
// is much more likely to be foldable here.
LargeOps.push_back(getSignExtendExpr(C, SrcType));
} else if (const SCEVMulExpr *M = dyn_cast<SCEVMulExpr>(Ops[i])) {
SmallVector<const SCEV *, 8> LargeMulOps;
for (unsigned j = 0, f = M->getNumOperands(); j != f && Ok; ++j) {
if (const SCEVTruncateExpr *T =
dyn_cast<SCEVTruncateExpr>(M->getOperand(j))) {
if (T->getOperand()->getType() != SrcType) {
Ok = false;
break;
}
LargeMulOps.push_back(T->getOperand());
} else if (const SCEVConstant *C =
dyn_cast<SCEVConstant>(M->getOperand(j))) {
// This could be either sign or zero extension, but sign extension
// is much more likely to be foldable here.
LargeMulOps.push_back(getSignExtendExpr(C, SrcType));
} else {
Ok = false;
break;
}
}
if (Ok)
LargeOps.push_back(getMulExpr(LargeMulOps));
} else {
Ok = false;
break;
}
}
if (Ok) {
// Evaluate the expression in the larger type.
const SCEV *Fold = getAddExpr(LargeOps);
// If it folds to something simple, use it. Otherwise, don't.
if (isa<SCEVConstant>(Fold) || isa<SCEVUnknown>(Fold))
return getTruncateExpr(Fold, DstType);
}
}
// Skip past any other cast SCEVs.
while (Idx < Ops.size() && Ops[Idx]->getSCEVType() < scAddExpr)
++Idx;
// If there are add operands they would be next.
if (Idx < Ops.size()) {
bool DeletedAdd = false;
while (const SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(Ops[Idx])) {
// If we have an add, expand the add operands onto the end of the operands
// list.
Ops.insert(Ops.end(), Add->op_begin(), Add->op_end());
Ops.erase(Ops.begin()+Idx);
DeletedAdd = true;
}
// If we deleted at least one add, we added operands to the end of the list,
// and they are not necessarily sorted. Recurse to resort and resimplify
// any operands we just aquired.
if (DeletedAdd)
return getAddExpr(Ops);
}
// Skip over the add expression until we get to a multiply.
while (Idx < Ops.size() && Ops[Idx]->getSCEVType() < scMulExpr)
++Idx;
// Check to see if there are any folding opportunities present with
// operands multiplied by constant values.
if (Idx < Ops.size() && isa<SCEVMulExpr>(Ops[Idx])) {
uint64_t BitWidth = getTypeSizeInBits(Ty);
DenseMap<const SCEV *, APInt> M;
SmallVector<const SCEV *, 8> NewOps;
APInt AccumulatedConstant(BitWidth, 0);
if (CollectAddOperandsWithScales(M, NewOps, AccumulatedConstant,
Ops, APInt(BitWidth, 1), *this)) {
// Some interesting folding opportunity is present, so its worthwhile to
// re-generate the operands list. Group the operands by constant scale,
// to avoid multiplying by the same constant scale multiple times.
std::map<APInt, SmallVector<const SCEV *, 4>, APIntCompare> MulOpLists;
for (SmallVector<const SCEV *, 8>::iterator I = NewOps.begin(),
E = NewOps.end(); I != E; ++I)
MulOpLists[M.find(*I)->second].push_back(*I);
// Re-generate the operands list.
Ops.clear();
if (AccumulatedConstant != 0)
Ops.push_back(getConstant(AccumulatedConstant));
for (std::map<APInt, SmallVector<const SCEV *, 4>, APIntCompare>::iterator
I = MulOpLists.begin(), E = MulOpLists.end(); I != E; ++I)
if (I->first != 0)
Ops.push_back(getMulExpr(getConstant(I->first),
getAddExpr(I->second)));
if (Ops.empty())
return getIntegerSCEV(0, Ty);
if (Ops.size() == 1)
return Ops[0];
return getAddExpr(Ops);
}
}
// If we are adding something to a multiply expression, make sure the
// something is not already an operand of the multiply. If so, merge it into
// the multiply.
for (; Idx < Ops.size() && isa<SCEVMulExpr>(Ops[Idx]); ++Idx) {
const SCEVMulExpr *Mul = cast<SCEVMulExpr>(Ops[Idx]);
for (unsigned MulOp = 0, e = Mul->getNumOperands(); MulOp != e; ++MulOp) {
const SCEV *MulOpSCEV = Mul->getOperand(MulOp);
for (unsigned AddOp = 0, e = Ops.size(); AddOp != e; ++AddOp)
if (MulOpSCEV == Ops[AddOp] && !isa<SCEVConstant>(Ops[AddOp])) {
// Fold W + X + (X * Y * Z) --> W + (X * ((Y*Z)+1))
const SCEV *InnerMul = Mul->getOperand(MulOp == 0);
if (Mul->getNumOperands() != 2) {
// If the multiply has more than two operands, we must get the
// Y*Z term.
SmallVector<const SCEV *, 4> MulOps(Mul->op_begin(), Mul->op_end());
MulOps.erase(MulOps.begin()+MulOp);
InnerMul = getMulExpr(MulOps);
}
const SCEV *One = getIntegerSCEV(1, Ty);
const SCEV *AddOne = getAddExpr(InnerMul, One);
const SCEV *OuterMul = getMulExpr(AddOne, Ops[AddOp]);
if (Ops.size() == 2) return OuterMul;
if (AddOp < Idx) {
Ops.erase(Ops.begin()+AddOp);
Ops.erase(Ops.begin()+Idx-1);
} else {
Ops.erase(Ops.begin()+Idx);
Ops.erase(Ops.begin()+AddOp-1);
}
Ops.push_back(OuterMul);
return getAddExpr(Ops);
}
// Check this multiply against other multiplies being added together.
for (unsigned OtherMulIdx = Idx+1;
OtherMulIdx < Ops.size() && isa<SCEVMulExpr>(Ops[OtherMulIdx]);
++OtherMulIdx) {
const SCEVMulExpr *OtherMul = cast<SCEVMulExpr>(Ops[OtherMulIdx]);
// If MulOp occurs in OtherMul, we can fold the two multiplies
// together.
for (unsigned OMulOp = 0, e = OtherMul->getNumOperands();
OMulOp != e; ++OMulOp)
if (OtherMul->getOperand(OMulOp) == MulOpSCEV) {
// Fold X + (A*B*C) + (A*D*E) --> X + (A*(B*C+D*E))
const SCEV *InnerMul1 = Mul->getOperand(MulOp == 0);
if (Mul->getNumOperands() != 2) {
SmallVector<const SCEV *, 4> MulOps(Mul->op_begin(),
Mul->op_end());
MulOps.erase(MulOps.begin()+MulOp);
InnerMul1 = getMulExpr(MulOps);
}
const SCEV *InnerMul2 = OtherMul->getOperand(OMulOp == 0);
if (OtherMul->getNumOperands() != 2) {
SmallVector<const SCEV *, 4> MulOps(OtherMul->op_begin(),
OtherMul->op_end());
MulOps.erase(MulOps.begin()+OMulOp);
InnerMul2 = getMulExpr(MulOps);
}
const SCEV *InnerMulSum = getAddExpr(InnerMul1,InnerMul2);
const SCEV *OuterMul = getMulExpr(MulOpSCEV, InnerMulSum);
if (Ops.size() == 2) return OuterMul;
Ops.erase(Ops.begin()+Idx);
Ops.erase(Ops.begin()+OtherMulIdx-1);
Ops.push_back(OuterMul);
return getAddExpr(Ops);
}
}
}
}
// If there are any add recurrences in the operands list, see if any other
// added values are loop invariant. If so, we can fold them into the
// recurrence.
while (Idx < Ops.size() && Ops[Idx]->getSCEVType() < scAddRecExpr)
++Idx;
// Scan over all recurrences, trying to fold loop invariants into them.
for (; Idx < Ops.size() && isa<SCEVAddRecExpr>(Ops[Idx]); ++Idx) {
// Scan all of the other operands to this add and add them to the vector if
// they are loop invariant w.r.t. the recurrence.
SmallVector<const SCEV *, 8> LIOps;
const SCEVAddRecExpr *AddRec = cast<SCEVAddRecExpr>(Ops[Idx]);
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
if (Ops[i]->isLoopInvariant(AddRec->getLoop())) {
LIOps.push_back(Ops[i]);
Ops.erase(Ops.begin()+i);
--i; --e;
}
// If we found some loop invariants, fold them into the recurrence.
if (!LIOps.empty()) {
// NLI + LI + {Start,+,Step} --> NLI + {LI+Start,+,Step}
LIOps.push_back(AddRec->getStart());
SmallVector<const SCEV *, 4> AddRecOps(AddRec->op_begin(),
AddRec->op_end());
AddRecOps[0] = getAddExpr(LIOps);
const SCEV *NewRec = getAddRecExpr(AddRecOps, AddRec->getLoop());
// If all of the other operands were loop invariant, we are done.
if (Ops.size() == 1) return NewRec;
// Otherwise, add the folded AddRec by the non-liv parts.
for (unsigned i = 0;; ++i)
if (Ops[i] == AddRec) {
Ops[i] = NewRec;
break;
}
return getAddExpr(Ops);
}
// Okay, if there weren't any loop invariants to be folded, check to see if
// there are multiple AddRec's with the same loop induction variable being
// added together. If so, we can fold them.
for (unsigned OtherIdx = Idx+1;
OtherIdx < Ops.size() && isa<SCEVAddRecExpr>(Ops[OtherIdx]);++OtherIdx)
if (OtherIdx != Idx) {
const SCEVAddRecExpr *OtherAddRec = cast<SCEVAddRecExpr>(Ops[OtherIdx]);
if (AddRec->getLoop() == OtherAddRec->getLoop()) {
// Other + {A,+,B} + {C,+,D} --> Other + {A+C,+,B+D}
SmallVector<const SCEV *, 4> NewOps(AddRec->op_begin(),
AddRec->op_end());
for (unsigned i = 0, e = OtherAddRec->getNumOperands(); i != e; ++i) {
if (i >= NewOps.size()) {
NewOps.insert(NewOps.end(), OtherAddRec->op_begin()+i,
OtherAddRec->op_end());
break;
}
NewOps[i] = getAddExpr(NewOps[i], OtherAddRec->getOperand(i));
}
const SCEV *NewAddRec = getAddRecExpr(NewOps, AddRec->getLoop());
if (Ops.size() == 2) return NewAddRec;
Ops.erase(Ops.begin()+Idx);
Ops.erase(Ops.begin()+OtherIdx-1);
Ops.push_back(NewAddRec);
return getAddExpr(Ops);
}
}
// Otherwise couldn't fold anything into this recurrence. Move onto the
// next one.
}
// Okay, it looks like we really DO need an add expr. Check to see if we
// already have one, otherwise create a new one.
FoldingSetNodeID ID;
ID.AddInteger(scAddExpr);
ID.AddInteger(Ops.size());
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
ID.AddPointer(Ops[i]);
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVAddExpr>();
new (S) SCEVAddExpr(ID, Ops);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
/// getMulExpr - Get a canonical multiply expression, or something simpler if
/// possible.
const SCEV *ScalarEvolution::getMulExpr(SmallVectorImpl<const SCEV *> &Ops) {
assert(!Ops.empty() && "Cannot get empty mul!");
#ifndef NDEBUG
for (unsigned i = 1, e = Ops.size(); i != e; ++i)
assert(getEffectiveSCEVType(Ops[i]->getType()) ==
getEffectiveSCEVType(Ops[0]->getType()) &&
"SCEVMulExpr operand types don't match!");
#endif
// Sort by complexity, this groups all similar expression types together.
GroupByComplexity(Ops, LI);
// If there are any constants, fold them together.
unsigned Idx = 0;
if (const SCEVConstant *LHSC = dyn_cast<SCEVConstant>(Ops[0])) {
// C1*(C2+V) -> C1*C2 + C1*V
if (Ops.size() == 2)
if (const SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(Ops[1]))
if (Add->getNumOperands() == 2 &&
isa<SCEVConstant>(Add->getOperand(0)))
return getAddExpr(getMulExpr(LHSC, Add->getOperand(0)),
getMulExpr(LHSC, Add->getOperand(1)));
++Idx;
while (const SCEVConstant *RHSC = dyn_cast<SCEVConstant>(Ops[Idx])) {
// We found two constants, fold them together!
ConstantInt *Fold = ConstantInt::get(getContext(),
LHSC->getValue()->getValue() *
RHSC->getValue()->getValue());
Ops[0] = getConstant(Fold);
Ops.erase(Ops.begin()+1); // Erase the folded element
if (Ops.size() == 1) return Ops[0];
LHSC = cast<SCEVConstant>(Ops[0]);
}
// If we are left with a constant one being multiplied, strip it off.
if (cast<SCEVConstant>(Ops[0])->getValue()->equalsInt(1)) {
Ops.erase(Ops.begin());
--Idx;
} else if (cast<SCEVConstant>(Ops[0])->getValue()->isZero()) {
// If we have a multiply of zero, it will always be zero.
return Ops[0];
}
}
// Skip over the add expression until we get to a multiply.
while (Idx < Ops.size() && Ops[Idx]->getSCEVType() < scMulExpr)
++Idx;
if (Ops.size() == 1)
return Ops[0];
// If there are mul operands inline them all into this expression.
if (Idx < Ops.size()) {
bool DeletedMul = false;
while (const SCEVMulExpr *Mul = dyn_cast<SCEVMulExpr>(Ops[Idx])) {
// If we have an mul, expand the mul operands onto the end of the operands
// list.
Ops.insert(Ops.end(), Mul->op_begin(), Mul->op_end());
Ops.erase(Ops.begin()+Idx);
DeletedMul = true;
}
// If we deleted at least one mul, we added operands to the end of the list,
// and they are not necessarily sorted. Recurse to resort and resimplify
// any operands we just aquired.
if (DeletedMul)
return getMulExpr(Ops);
}
// If there are any add recurrences in the operands list, see if any other
// added values are loop invariant. If so, we can fold them into the
// recurrence.
while (Idx < Ops.size() && Ops[Idx]->getSCEVType() < scAddRecExpr)
++Idx;
// Scan over all recurrences, trying to fold loop invariants into them.
for (; Idx < Ops.size() && isa<SCEVAddRecExpr>(Ops[Idx]); ++Idx) {
// Scan all of the other operands to this mul and add them to the vector if
// they are loop invariant w.r.t. the recurrence.
SmallVector<const SCEV *, 8> LIOps;
const SCEVAddRecExpr *AddRec = cast<SCEVAddRecExpr>(Ops[Idx]);
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
if (Ops[i]->isLoopInvariant(AddRec->getLoop())) {
LIOps.push_back(Ops[i]);
Ops.erase(Ops.begin()+i);
--i; --e;
}
// If we found some loop invariants, fold them into the recurrence.
if (!LIOps.empty()) {
// NLI * LI * {Start,+,Step} --> NLI * {LI*Start,+,LI*Step}
SmallVector<const SCEV *, 4> NewOps;
NewOps.reserve(AddRec->getNumOperands());
if (LIOps.size() == 1) {
const SCEV *Scale = LIOps[0];
for (unsigned i = 0, e = AddRec->getNumOperands(); i != e; ++i)
NewOps.push_back(getMulExpr(Scale, AddRec->getOperand(i)));
} else {
for (unsigned i = 0, e = AddRec->getNumOperands(); i != e; ++i) {
SmallVector<const SCEV *, 4> MulOps(LIOps.begin(), LIOps.end());
MulOps.push_back(AddRec->getOperand(i));
NewOps.push_back(getMulExpr(MulOps));
}
}
const SCEV *NewRec = getAddRecExpr(NewOps, AddRec->getLoop());
// If all of the other operands were loop invariant, we are done.
if (Ops.size() == 1) return NewRec;
// Otherwise, multiply the folded AddRec by the non-liv parts.
for (unsigned i = 0;; ++i)
if (Ops[i] == AddRec) {
Ops[i] = NewRec;
break;
}
return getMulExpr(Ops);
}
// Okay, if there weren't any loop invariants to be folded, check to see if
// there are multiple AddRec's with the same loop induction variable being
// multiplied together. If so, we can fold them.
for (unsigned OtherIdx = Idx+1;
OtherIdx < Ops.size() && isa<SCEVAddRecExpr>(Ops[OtherIdx]);++OtherIdx)
if (OtherIdx != Idx) {
const SCEVAddRecExpr *OtherAddRec = cast<SCEVAddRecExpr>(Ops[OtherIdx]);
if (AddRec->getLoop() == OtherAddRec->getLoop()) {
// F * G --> {A,+,B} * {C,+,D} --> {A*C,+,F*D + G*B + B*D}
const SCEVAddRecExpr *F = AddRec, *G = OtherAddRec;
const SCEV *NewStart = getMulExpr(F->getStart(),
G->getStart());
const SCEV *B = F->getStepRecurrence(*this);
const SCEV *D = G->getStepRecurrence(*this);
const SCEV *NewStep = getAddExpr(getMulExpr(F, D),
getMulExpr(G, B),
getMulExpr(B, D));
const SCEV *NewAddRec = getAddRecExpr(NewStart, NewStep,
F->getLoop());
if (Ops.size() == 2) return NewAddRec;
Ops.erase(Ops.begin()+Idx);
Ops.erase(Ops.begin()+OtherIdx-1);
Ops.push_back(NewAddRec);
return getMulExpr(Ops);
}
}
// Otherwise couldn't fold anything into this recurrence. Move onto the
// next one.
}
// Okay, it looks like we really DO need an mul expr. Check to see if we
// already have one, otherwise create a new one.
FoldingSetNodeID ID;
ID.AddInteger(scMulExpr);
ID.AddInteger(Ops.size());
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
ID.AddPointer(Ops[i]);
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVMulExpr>();
new (S) SCEVMulExpr(ID, Ops);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
/// getUDivExpr - Get a canonical unsigned division expression, or something
/// simpler if possible.
const SCEV *ScalarEvolution::getUDivExpr(const SCEV *LHS,
const SCEV *RHS) {
assert(getEffectiveSCEVType(LHS->getType()) ==
getEffectiveSCEVType(RHS->getType()) &&
"SCEVUDivExpr operand types don't match!");
if (const SCEVConstant *RHSC = dyn_cast<SCEVConstant>(RHS)) {
if (RHSC->getValue()->equalsInt(1))
return LHS; // X udiv 1 --> x
if (RHSC->isZero())
return getIntegerSCEV(0, LHS->getType()); // value is undefined
// Determine if the division can be folded into the operands of
// its operands.
// TODO: Generalize this to non-constants by using known-bits information.
const Type *Ty = LHS->getType();
unsigned LZ = RHSC->getValue()->getValue().countLeadingZeros();
unsigned MaxShiftAmt = getTypeSizeInBits(Ty) - LZ;
// For non-power-of-two values, effectively round the value up to the
// nearest power of two.
if (!RHSC->getValue()->getValue().isPowerOf2())
++MaxShiftAmt;
const IntegerType *ExtTy =
IntegerType::get(getContext(), getTypeSizeInBits(Ty) + MaxShiftAmt);
// {X,+,N}/C --> {X/C,+,N/C} if safe and N/C can be folded.
if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(LHS))
if (const SCEVConstant *Step =
dyn_cast<SCEVConstant>(AR->getStepRecurrence(*this)))
if (!Step->getValue()->getValue()
.urem(RHSC->getValue()->getValue()) &&
getZeroExtendExpr(AR, ExtTy) ==
getAddRecExpr(getZeroExtendExpr(AR->getStart(), ExtTy),
getZeroExtendExpr(Step, ExtTy),
AR->getLoop())) {
SmallVector<const SCEV *, 4> Operands;
for (unsigned i = 0, e = AR->getNumOperands(); i != e; ++i)
Operands.push_back(getUDivExpr(AR->getOperand(i), RHS));
return getAddRecExpr(Operands, AR->getLoop());
}
// (A*B)/C --> A*(B/C) if safe and B/C can be folded.
if (const SCEVMulExpr *M = dyn_cast<SCEVMulExpr>(LHS)) {
SmallVector<const SCEV *, 4> Operands;
for (unsigned i = 0, e = M->getNumOperands(); i != e; ++i)
Operands.push_back(getZeroExtendExpr(M->getOperand(i), ExtTy));
if (getZeroExtendExpr(M, ExtTy) == getMulExpr(Operands))
// Find an operand that's safely divisible.
for (unsigned i = 0, e = M->getNumOperands(); i != e; ++i) {
const SCEV *Op = M->getOperand(i);
const SCEV *Div = getUDivExpr(Op, RHSC);
if (!isa<SCEVUDivExpr>(Div) && getMulExpr(Div, RHSC) == Op) {
const SmallVectorImpl<const SCEV *> &MOperands = M->getOperands();
Operands = SmallVector<const SCEV *, 4>(MOperands.begin(),
MOperands.end());
Operands[i] = Div;
return getMulExpr(Operands);
}
}
}
// (A+B)/C --> (A/C + B/C) if safe and A/C and B/C can be folded.
if (const SCEVAddRecExpr *A = dyn_cast<SCEVAddRecExpr>(LHS)) {
SmallVector<const SCEV *, 4> Operands;
for (unsigned i = 0, e = A->getNumOperands(); i != e; ++i)
Operands.push_back(getZeroExtendExpr(A->getOperand(i), ExtTy));
if (getZeroExtendExpr(A, ExtTy) == getAddExpr(Operands)) {
Operands.clear();
for (unsigned i = 0, e = A->getNumOperands(); i != e; ++i) {
const SCEV *Op = getUDivExpr(A->getOperand(i), RHS);
if (isa<SCEVUDivExpr>(Op) || getMulExpr(Op, RHS) != A->getOperand(i))
break;
Operands.push_back(Op);
}
if (Operands.size() == A->getNumOperands())
return getAddExpr(Operands);
}
}
// Fold if both operands are constant.
if (const SCEVConstant *LHSC = dyn_cast<SCEVConstant>(LHS)) {
Constant *LHSCV = LHSC->getValue();
Constant *RHSCV = RHSC->getValue();
return getConstant(cast<ConstantInt>(ConstantExpr::getUDiv(LHSCV,
RHSCV)));
}
}
FoldingSetNodeID ID;
ID.AddInteger(scUDivExpr);
ID.AddPointer(LHS);
ID.AddPointer(RHS);
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVUDivExpr>();
new (S) SCEVUDivExpr(ID, LHS, RHS);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
/// getAddRecExpr - Get an add recurrence expression for the specified loop.
/// Simplify the expression as much as possible.
const SCEV *ScalarEvolution::getAddRecExpr(const SCEV *Start,
const SCEV *Step, const Loop *L) {
SmallVector<const SCEV *, 4> Operands;
Operands.push_back(Start);
if (const SCEVAddRecExpr *StepChrec = dyn_cast<SCEVAddRecExpr>(Step))
if (StepChrec->getLoop() == L) {
Operands.insert(Operands.end(), StepChrec->op_begin(),
StepChrec->op_end());
return getAddRecExpr(Operands, L);
}
Operands.push_back(Step);
return getAddRecExpr(Operands, L);
}
/// getAddRecExpr - Get an add recurrence expression for the specified loop.
/// Simplify the expression as much as possible.
const SCEV *
ScalarEvolution::getAddRecExpr(SmallVectorImpl<const SCEV *> &Operands,
const Loop *L) {
if (Operands.size() == 1) return Operands[0];
#ifndef NDEBUG
for (unsigned i = 1, e = Operands.size(); i != e; ++i)
assert(getEffectiveSCEVType(Operands[i]->getType()) ==
getEffectiveSCEVType(Operands[0]->getType()) &&
"SCEVAddRecExpr operand types don't match!");
#endif
if (Operands.back()->isZero()) {
Operands.pop_back();
return getAddRecExpr(Operands, L); // {X,+,0} --> X
}
// Canonicalize nested AddRecs in by nesting them in order of loop depth.
if (const SCEVAddRecExpr *NestedAR = dyn_cast<SCEVAddRecExpr>(Operands[0])) {
const Loop* NestedLoop = NestedAR->getLoop();
if (L->getLoopDepth() < NestedLoop->getLoopDepth()) {
SmallVector<const SCEV *, 4> NestedOperands(NestedAR->op_begin(),
NestedAR->op_end());
Operands[0] = NestedAR->getStart();
// AddRecs require their operands be loop-invariant with respect to their
// loops. Don't perform this transformation if it would break this
// requirement.
bool AllInvariant = true;
for (unsigned i = 0, e = Operands.size(); i != e; ++i)
if (!Operands[i]->isLoopInvariant(L)) {
AllInvariant = false;
break;
}
if (AllInvariant) {
NestedOperands[0] = getAddRecExpr(Operands, L);
AllInvariant = true;
for (unsigned i = 0, e = NestedOperands.size(); i != e; ++i)
if (!NestedOperands[i]->isLoopInvariant(NestedLoop)) {
AllInvariant = false;
break;
}
if (AllInvariant)
// Ok, both add recurrences are valid after the transformation.
return getAddRecExpr(NestedOperands, NestedLoop);
}
// Reset Operands to its original state.
Operands[0] = NestedAR;
}
}
FoldingSetNodeID ID;
ID.AddInteger(scAddRecExpr);
ID.AddInteger(Operands.size());
for (unsigned i = 0, e = Operands.size(); i != e; ++i)
ID.AddPointer(Operands[i]);
ID.AddPointer(L);
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVAddRecExpr>();
new (S) SCEVAddRecExpr(ID, Operands, L);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
const SCEV *ScalarEvolution::getSMaxExpr(const SCEV *LHS,
const SCEV *RHS) {
SmallVector<const SCEV *, 2> Ops;
Ops.push_back(LHS);
Ops.push_back(RHS);
return getSMaxExpr(Ops);
}
const SCEV *
ScalarEvolution::getSMaxExpr(SmallVectorImpl<const SCEV *> &Ops) {
assert(!Ops.empty() && "Cannot get empty smax!");
if (Ops.size() == 1) return Ops[0];
#ifndef NDEBUG
for (unsigned i = 1, e = Ops.size(); i != e; ++i)
assert(getEffectiveSCEVType(Ops[i]->getType()) ==
getEffectiveSCEVType(Ops[0]->getType()) &&
"SCEVSMaxExpr operand types don't match!");
#endif
// Sort by complexity, this groups all similar expression types together.
GroupByComplexity(Ops, LI);
// If there are any constants, fold them together.
unsigned Idx = 0;
if (const SCEVConstant *LHSC = dyn_cast<SCEVConstant>(Ops[0])) {
++Idx;
assert(Idx < Ops.size());
while (const SCEVConstant *RHSC = dyn_cast<SCEVConstant>(Ops[Idx])) {
// We found two constants, fold them together!
ConstantInt *Fold = ConstantInt::get(getContext(),
APIntOps::smax(LHSC->getValue()->getValue(),
RHSC->getValue()->getValue()));
Ops[0] = getConstant(Fold);
Ops.erase(Ops.begin()+1); // Erase the folded element
if (Ops.size() == 1) return Ops[0];
LHSC = cast<SCEVConstant>(Ops[0]);
}
// If we are left with a constant minimum-int, strip it off.
if (cast<SCEVConstant>(Ops[0])->getValue()->isMinValue(true)) {
Ops.erase(Ops.begin());
--Idx;
} else if (cast<SCEVConstant>(Ops[0])->getValue()->isMaxValue(true)) {
// If we have an smax with a constant maximum-int, it will always be
// maximum-int.
return Ops[0];
}
}
if (Ops.size() == 1) return Ops[0];
// Find the first SMax
while (Idx < Ops.size() && Ops[Idx]->getSCEVType() < scSMaxExpr)
++Idx;
// Check to see if one of the operands is an SMax. If so, expand its operands
// onto our operand list, and recurse to simplify.
if (Idx < Ops.size()) {
bool DeletedSMax = false;
while (const SCEVSMaxExpr *SMax = dyn_cast<SCEVSMaxExpr>(Ops[Idx])) {
Ops.insert(Ops.end(), SMax->op_begin(), SMax->op_end());
Ops.erase(Ops.begin()+Idx);
DeletedSMax = true;
}
if (DeletedSMax)
return getSMaxExpr(Ops);
}
// Okay, check to see if the same value occurs in the operand list twice. If
// so, delete one. Since we sorted the list, these values are required to
// be adjacent.
for (unsigned i = 0, e = Ops.size()-1; i != e; ++i)
if (Ops[i] == Ops[i+1]) { // X smax Y smax Y --> X smax Y
Ops.erase(Ops.begin()+i, Ops.begin()+i+1);
--i; --e;
}
if (Ops.size() == 1) return Ops[0];
assert(!Ops.empty() && "Reduced smax down to nothing!");
// Okay, it looks like we really DO need an smax expr. Check to see if we
// already have one, otherwise create a new one.
FoldingSetNodeID ID;
ID.AddInteger(scSMaxExpr);
ID.AddInteger(Ops.size());
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
ID.AddPointer(Ops[i]);
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVSMaxExpr>();
new (S) SCEVSMaxExpr(ID, Ops);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
const SCEV *ScalarEvolution::getUMaxExpr(const SCEV *LHS,
const SCEV *RHS) {
SmallVector<const SCEV *, 2> Ops;
Ops.push_back(LHS);
Ops.push_back(RHS);
return getUMaxExpr(Ops);
}
const SCEV *
ScalarEvolution::getUMaxExpr(SmallVectorImpl<const SCEV *> &Ops) {
assert(!Ops.empty() && "Cannot get empty umax!");
if (Ops.size() == 1) return Ops[0];
#ifndef NDEBUG
for (unsigned i = 1, e = Ops.size(); i != e; ++i)
assert(getEffectiveSCEVType(Ops[i]->getType()) ==
getEffectiveSCEVType(Ops[0]->getType()) &&
"SCEVUMaxExpr operand types don't match!");
#endif
// Sort by complexity, this groups all similar expression types together.
GroupByComplexity(Ops, LI);
// If there are any constants, fold them together.
unsigned Idx = 0;
if (const SCEVConstant *LHSC = dyn_cast<SCEVConstant>(Ops[0])) {
++Idx;
assert(Idx < Ops.size());
while (const SCEVConstant *RHSC = dyn_cast<SCEVConstant>(Ops[Idx])) {
// We found two constants, fold them together!
ConstantInt *Fold = ConstantInt::get(getContext(),
APIntOps::umax(LHSC->getValue()->getValue(),
RHSC->getValue()->getValue()));
Ops[0] = getConstant(Fold);
Ops.erase(Ops.begin()+1); // Erase the folded element
if (Ops.size() == 1) return Ops[0];
LHSC = cast<SCEVConstant>(Ops[0]);
}
// If we are left with a constant minimum-int, strip it off.
if (cast<SCEVConstant>(Ops[0])->getValue()->isMinValue(false)) {
Ops.erase(Ops.begin());
--Idx;
} else if (cast<SCEVConstant>(Ops[0])->getValue()->isMaxValue(false)) {
// If we have an umax with a constant maximum-int, it will always be
// maximum-int.
return Ops[0];
}
}
if (Ops.size() == 1) return Ops[0];
// Find the first UMax
while (Idx < Ops.size() && Ops[Idx]->getSCEVType() < scUMaxExpr)
++Idx;
// Check to see if one of the operands is a UMax. If so, expand its operands
// onto our operand list, and recurse to simplify.
if (Idx < Ops.size()) {
bool DeletedUMax = false;
while (const SCEVUMaxExpr *UMax = dyn_cast<SCEVUMaxExpr>(Ops[Idx])) {
Ops.insert(Ops.end(), UMax->op_begin(), UMax->op_end());
Ops.erase(Ops.begin()+Idx);
DeletedUMax = true;
}
if (DeletedUMax)
return getUMaxExpr(Ops);
}
// Okay, check to see if the same value occurs in the operand list twice. If
// so, delete one. Since we sorted the list, these values are required to
// be adjacent.
for (unsigned i = 0, e = Ops.size()-1; i != e; ++i)
if (Ops[i] == Ops[i+1]) { // X umax Y umax Y --> X umax Y
Ops.erase(Ops.begin()+i, Ops.begin()+i+1);
--i; --e;
}
if (Ops.size() == 1) return Ops[0];
assert(!Ops.empty() && "Reduced umax down to nothing!");
// Okay, it looks like we really DO need a umax expr. Check to see if we
// already have one, otherwise create a new one.
FoldingSetNodeID ID;
ID.AddInteger(scUMaxExpr);
ID.AddInteger(Ops.size());
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
ID.AddPointer(Ops[i]);
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVUMaxExpr>();
new (S) SCEVUMaxExpr(ID, Ops);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
const SCEV *ScalarEvolution::getSMinExpr(const SCEV *LHS,
const SCEV *RHS) {
// ~smax(~x, ~y) == smin(x, y).
return getNotSCEV(getSMaxExpr(getNotSCEV(LHS), getNotSCEV(RHS)));
}
const SCEV *ScalarEvolution::getUMinExpr(const SCEV *LHS,
const SCEV *RHS) {
// ~umax(~x, ~y) == umin(x, y)
return getNotSCEV(getUMaxExpr(getNotSCEV(LHS), getNotSCEV(RHS)));
}
const SCEV *ScalarEvolution::getFieldOffsetExpr(const StructType *STy,
unsigned FieldNo) {
// If we have TargetData we can determine the constant offset.
if (TD) {
const Type *IntPtrTy = TD->getIntPtrType(getContext());
const StructLayout &SL = *TD->getStructLayout(STy);
uint64_t Offset = SL.getElementOffset(FieldNo);
return getIntegerSCEV(Offset, IntPtrTy);
}
// Field 0 is always at offset 0.
if (FieldNo == 0) {
const Type *Ty = getEffectiveSCEVType(PointerType::getUnqual(STy));
return getIntegerSCEV(0, Ty);
}
// Okay, it looks like we really DO need an offsetof expr. Check to see if we
// already have one, otherwise create a new one.
FoldingSetNodeID ID;
ID.AddInteger(scFieldOffset);
ID.AddPointer(STy);
ID.AddInteger(FieldNo);
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVFieldOffsetExpr>();
const Type *Ty = getEffectiveSCEVType(PointerType::getUnqual(STy));
new (S) SCEVFieldOffsetExpr(ID, Ty, STy, FieldNo);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
const SCEV *ScalarEvolution::getAllocSizeExpr(const Type *AllocTy) {
// If we have TargetData we can determine the constant size.
if (TD && AllocTy->isSized()) {
const Type *IntPtrTy = TD->getIntPtrType(getContext());
return getIntegerSCEV(TD->getTypeAllocSize(AllocTy), IntPtrTy);
}
// Expand an array size into the element size times the number
// of elements.
if (const ArrayType *ATy = dyn_cast<ArrayType>(AllocTy)) {
const SCEV *E = getAllocSizeExpr(ATy->getElementType());
return getMulExpr(
E, getConstant(ConstantInt::get(cast<IntegerType>(E->getType()),
ATy->getNumElements())));
}
// Expand a vector size into the element size times the number
// of elements.
if (const VectorType *VTy = dyn_cast<VectorType>(AllocTy)) {
const SCEV *E = getAllocSizeExpr(VTy->getElementType());
return getMulExpr(
E, getConstant(ConstantInt::get(cast<IntegerType>(E->getType()),
VTy->getNumElements())));
}
// Okay, it looks like we really DO need a sizeof expr. Check to see if we
// already have one, otherwise create a new one.
FoldingSetNodeID ID;
ID.AddInteger(scAllocSize);
ID.AddPointer(AllocTy);
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVAllocSizeExpr>();
const Type *Ty = getEffectiveSCEVType(PointerType::getUnqual(AllocTy));
new (S) SCEVAllocSizeExpr(ID, Ty, AllocTy);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
const SCEV *ScalarEvolution::getUnknown(Value *V) {
// Don't attempt to do anything other than create a SCEVUnknown object
// here. createSCEV only calls getUnknown after checking for all other
// interesting possibilities, and any other code that calls getUnknown
// is doing so in order to hide a value from SCEV canonicalization.
FoldingSetNodeID ID;
ID.AddInteger(scUnknown);
ID.AddPointer(V);
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
SCEV *S = SCEVAllocator.Allocate<SCEVUnknown>();
new (S) SCEVUnknown(ID, V);
UniqueSCEVs.InsertNode(S, IP);
return S;
}
//===----------------------------------------------------------------------===//
// Basic SCEV Analysis and PHI Idiom Recognition Code
//
/// isSCEVable - Test if values of the given type are analyzable within
/// the SCEV framework. This primarily includes integer types, and it
/// can optionally include pointer types if the ScalarEvolution class
/// has access to target-specific information.
bool ScalarEvolution::isSCEVable(const Type *Ty) const {
// Integers and pointers are always SCEVable.
return Ty->isInteger() || isa<PointerType>(Ty);
}
/// getTypeSizeInBits - Return the size in bits of the specified type,
/// for which isSCEVable must return true.
uint64_t ScalarEvolution::getTypeSizeInBits(const Type *Ty) const {
assert(isSCEVable(Ty) && "Type is not SCEVable!");
// If we have a TargetData, use it!
if (TD)
return TD->getTypeSizeInBits(Ty);
// Integer types have fixed sizes.
if (Ty->isInteger())
return Ty->getPrimitiveSizeInBits();
// The only other support type is pointer. Without TargetData, conservatively
// assume pointers are 64-bit.
assert(isa<PointerType>(Ty) && "isSCEVable permitted a non-SCEVable type!");
return 64;
}
/// getEffectiveSCEVType - Return a type with the same bitwidth as
/// the given type and which represents how SCEV will treat the given
/// type, for which isSCEVable must return true. For pointer types,
/// this is the pointer-sized integer type.
const Type *ScalarEvolution::getEffectiveSCEVType(const Type *Ty) const {
assert(isSCEVable(Ty) && "Type is not SCEVable!");
if (Ty->isInteger())
return Ty;
// The only other support type is pointer.
assert(isa<PointerType>(Ty) && "Unexpected non-pointer non-integer type!");
if (TD) return TD->getIntPtrType(getContext());
// Without TargetData, conservatively assume pointers are 64-bit.
return Type::getInt64Ty(getContext());
}
const SCEV *ScalarEvolution::getCouldNotCompute() {
return &CouldNotCompute;
}
/// getSCEV - Return an existing SCEV if it exists, otherwise analyze the
/// expression and create a new one.
const SCEV *ScalarEvolution::getSCEV(Value *V) {
assert(isSCEVable(V->getType()) && "Value is not SCEVable!");
std::map<SCEVCallbackVH, const SCEV *>::iterator I = Scalars.find(V);
if (I != Scalars.end()) return I->second;
const SCEV *S = createSCEV(V);
Scalars.insert(std::make_pair(SCEVCallbackVH(V, this), S));
return S;
}
/// getIntegerSCEV - Given a SCEVable type, create a constant for the
/// specified signed integer value and return a SCEV for the constant.
const SCEV *ScalarEvolution::getIntegerSCEV(int Val, const Type *Ty) {
const IntegerType *ITy = cast<IntegerType>(getEffectiveSCEVType(Ty));
return getConstant(ConstantInt::get(ITy, Val));
}
/// getNegativeSCEV - Return a SCEV corresponding to -V = -1*V
///
const SCEV *ScalarEvolution::getNegativeSCEV(const SCEV *V) {
if (const SCEVConstant *VC = dyn_cast<SCEVConstant>(V))
return getConstant(
cast<ConstantInt>(ConstantExpr::getNeg(VC->getValue())));
const Type *Ty = V->getType();
Ty = getEffectiveSCEVType(Ty);
return getMulExpr(V,
getConstant(cast<ConstantInt>(Constant::getAllOnesValue(Ty))));
}
/// getNotSCEV - Return a SCEV corresponding to ~V = -1-V
const SCEV *ScalarEvolution::getNotSCEV(const SCEV *V) {
if (const SCEVConstant *VC = dyn_cast<SCEVConstant>(V))
return getConstant(
cast<ConstantInt>(ConstantExpr::getNot(VC->getValue())));
const Type *Ty = V->getType();
Ty = getEffectiveSCEVType(Ty);
const SCEV *AllOnes =
getConstant(cast<ConstantInt>(Constant::getAllOnesValue(Ty)));
return getMinusSCEV(AllOnes, V);
}
/// getMinusSCEV - Return a SCEV corresponding to LHS - RHS.
///
const SCEV *ScalarEvolution::getMinusSCEV(const SCEV *LHS,
const SCEV *RHS) {
// X - Y --> X + -Y
return getAddExpr(LHS, getNegativeSCEV(RHS));
}
/// getTruncateOrZeroExtend - Return a SCEV corresponding to a conversion of the
/// input value to the specified type. If the type must be extended, it is zero
/// extended.
const SCEV *
ScalarEvolution::getTruncateOrZeroExtend(const SCEV *V,
const Type *Ty) {
const Type *SrcTy = V->getType();
assert((SrcTy->isInteger() || isa<PointerType>(SrcTy)) &&
(Ty->isInteger() || isa<PointerType>(Ty)) &&
"Cannot truncate or zero extend with non-integer arguments!");
if (getTypeSizeInBits(SrcTy) == getTypeSizeInBits(Ty))
return V; // No conversion
if (getTypeSizeInBits(SrcTy) > getTypeSizeInBits(Ty))
return getTruncateExpr(V, Ty);
return getZeroExtendExpr(V, Ty);
}
/// getTruncateOrSignExtend - Return a SCEV corresponding to a conversion of the
/// input value to the specified type. If the type must be extended, it is sign
/// extended.
const SCEV *
ScalarEvolution::getTruncateOrSignExtend(const SCEV *V,
const Type *Ty) {
const Type *SrcTy = V->getType();
assert((SrcTy->isInteger() || isa<PointerType>(SrcTy)) &&
(Ty->isInteger() || isa<PointerType>(Ty)) &&
"Cannot truncate or zero extend with non-integer arguments!");
if (getTypeSizeInBits(SrcTy) == getTypeSizeInBits(Ty))
return V; // No conversion
if (getTypeSizeInBits(SrcTy) > getTypeSizeInBits(Ty))
return getTruncateExpr(V, Ty);
return getSignExtendExpr(V, Ty);
}
/// getNoopOrZeroExtend - Return a SCEV corresponding to a conversion of the
/// input value to the specified type. If the type must be extended, it is zero
/// extended. The conversion must not be narrowing.
const SCEV *
ScalarEvolution::getNoopOrZeroExtend(const SCEV *V, const Type *Ty) {
const Type *SrcTy = V->getType();
assert((SrcTy->isInteger() || isa<PointerType>(SrcTy)) &&
(Ty->isInteger() || isa<PointerType>(Ty)) &&
"Cannot noop or zero extend with non-integer arguments!");
assert(getTypeSizeInBits(SrcTy) <= getTypeSizeInBits(Ty) &&
"getNoopOrZeroExtend cannot truncate!");
if (getTypeSizeInBits(SrcTy) == getTypeSizeInBits(Ty))
return V; // No conversion
return getZeroExtendExpr(V, Ty);
}
/// getNoopOrSignExtend - Return a SCEV corresponding to a conversion of the
/// input value to the specified type. If the type must be extended, it is sign
/// extended. The conversion must not be narrowing.
const SCEV *
ScalarEvolution::getNoopOrSignExtend(const SCEV *V, const Type *Ty) {
const Type *SrcTy = V->getType();
assert((SrcTy->isInteger() || isa<PointerType>(SrcTy)) &&
(Ty->isInteger() || isa<PointerType>(Ty)) &&
"Cannot noop or sign extend with non-integer arguments!");
assert(getTypeSizeInBits(SrcTy) <= getTypeSizeInBits(Ty) &&
"getNoopOrSignExtend cannot truncate!");
if (getTypeSizeInBits(SrcTy) == getTypeSizeInBits(Ty))
return V; // No conversion
return getSignExtendExpr(V, Ty);
}
/// getNoopOrAnyExtend - Return a SCEV corresponding to a conversion of
/// the input value to the specified type. If the type must be extended,
/// it is extended with unspecified bits. The conversion must not be
/// narrowing.
const SCEV *
ScalarEvolution::getNoopOrAnyExtend