#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/Assembly/Writer.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
#include "llvm/Target/TargetLibraryInfo.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/ConstantRange.h"
"derived loop"),
cl::init(100));
+// FIXME: Enable this with XDEBUG when the test suite is clean.
+static cl::opt<bool>
+VerifySCEV("verify-scev",
+ cl::desc("Verify ScalarEvolution's backedge taken counts (slow)"));
+
INITIALIZE_PASS_BEGIN(ScalarEvolution, "scalar-evolution",
"Scalar Evolution Analysis", false, true)
INITIALIZE_PASS_DEPENDENCY(LoopInfo)
// Implementation of the SCEV class.
//
+#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
void SCEV::dump() const {
print(dbgs());
dbgs() << '\n';
}
+#endif
void SCEV::print(raw_ostream &OS) const {
switch (getSCEVType()) {
return getAddRecExpr(Operands, AddRec->getLoop(), SCEV::FlagAnyWrap);
}
- // As a special case, fold trunc(undef) to undef. We don't want to
- // know too much about SCEVUnknowns, but this special case is handy
- // and harmless.
- if (const SCEVUnknown *U = dyn_cast<SCEVUnknown>(Op))
- if (isa<UndefValue>(U->getValue()))
- return getSCEV(UndefValue::get(Ty));
-
// The cast wasn't folded; create an explicit cast node. We can reuse
// the existing insert position since if we get here, we won't have
// made any changes which would invalidate it.
return getAddRecExpr(Ops, AR->getLoop(), SCEV::FlagNW);
}
- // As a special case, fold anyext(undef) to undef. We don't want to
- // know too much about SCEVUnknowns, but this special case is handy
- // and harmless.
- if (const SCEVUnknown *U = dyn_cast<SCEVUnknown>(Op))
- if (isa<UndefValue>(U->getValue()))
- return getSCEV(UndefValue::get(Ty));
-
// If the expression is obviously signed, use the sext cast value.
if (isa<SCEVSMaxExpr>(Op))
return SExt;
/// Compute the result of "n choose k", the binomial coefficient. If an
/// intermediate computation overflows, Overflow will be set and the return will
-/// be garbage. Overflow is not cleared on absense of overflow.
+/// be garbage. Overflow is not cleared on absence of overflow.
static uint64_t Choose(uint64_t n, uint64_t k, bool &Overflow) {
// We use the multiplicative formula:
// n(n-1)(n-2)...(n-(k-1)) / k(k-1)(k-2)...1 .
return getNotSCEV(getUMaxExpr(getNotSCEV(LHS), getNotSCEV(RHS)));
}
-const SCEV *ScalarEvolution::getSizeOfExpr(Type *AllocTy) {
- // If we have TargetData, we can bypass creating a target-independent
+const SCEV *ScalarEvolution::getSizeOfExpr(Type *AllocTy, Type *IntPtrTy) {
+ // If we have DataLayout, we can bypass creating a target-independent
// constant expression and then folding it back into a ConstantInt.
// This is just a compile-time optimization.
if (TD)
- return getConstant(TD->getIntPtrType(getContext()),
- TD->getTypeAllocSize(AllocTy));
+ return getConstant(IntPtrTy, TD->getTypeAllocSize(AllocTy));
Constant *C = ConstantExpr::getSizeOf(AllocTy);
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C))
return getTruncateOrZeroExtend(getSCEV(C), Ty);
}
-const SCEV *ScalarEvolution::getOffsetOfExpr(StructType *STy,
+const SCEV *ScalarEvolution::getOffsetOfExpr(StructType *STy, Type *IntPtrTy,
unsigned FieldNo) {
- // If we have TargetData, we can bypass creating a target-independent
+ // If we have DataLayout, we can bypass creating a target-independent
// constant expression and then folding it back into a ConstantInt.
// This is just a compile-time optimization.
if (TD)
- return getConstant(TD->getIntPtrType(getContext()),
+ return getConstant(IntPtrTy,
TD->getStructLayout(STy)->getElementOffset(FieldNo));
Constant *C = ConstantExpr::getOffsetOf(STy, FieldNo);
uint64_t ScalarEvolution::getTypeSizeInBits(Type *Ty) const {
assert(isSCEVable(Ty) && "Type is not SCEVable!");
- // If we have a TargetData, use it!
+ // If we have a DataLayout, use it!
if (TD)
return TD->getTypeSizeInBits(Ty);
if (Ty->isIntegerTy())
return Ty->getPrimitiveSizeInBits();
- // The only other support type is pointer. Without TargetData, conservatively
+ // The only other support type is pointer. Without DataLayout, conservatively
// assume pointers are 64-bit.
assert(Ty->isPointerTy() && "isSCEVable permitted a non-SCEVable type!");
return 64;
// The only other support type is pointer.
assert(Ty->isPointerTy() && "Unexpected non-pointer non-integer type!");
- if (TD) return TD->getIntPtrType(getContext());
+ if (TD) return TD->getIntPtrType(Ty);
- // Without TargetData, conservatively assume pointers are 64-bit.
+ // Without DataLayout, conservatively assume pointers are 64-bit.
return Type::getInt64Ty(getContext());
}
const SCEV *ScalarEvolution::getSCEV(Value *V) {
assert(isSCEVable(V->getType()) && "Value is not SCEVable!");
- ValueExprMapType::const_iterator I = ValueExprMap.find(V);
+ ValueExprMapType::const_iterator I = ValueExprMap.find_as(V);
if (I != ValueExprMap.end()) return I->second;
const SCEV *S = createSCEV(V);
if (!Visited.insert(I)) continue;
ValueExprMapType::iterator It =
- ValueExprMap.find(static_cast<Value *>(I));
+ ValueExprMap.find_as(static_cast<Value *>(I));
if (It != ValueExprMap.end()) {
const SCEV *Old = It->second;
if (BEValueV && StartValueV) {
// While we are analyzing this PHI node, handle its value symbolically.
const SCEV *SymbolicName = getUnknown(PN);
- assert(ValueExprMap.find(PN) == ValueExprMap.end() &&
+ assert(ValueExprMap.find_as(PN) == ValueExprMap.end() &&
"PHI node already processed?");
ValueExprMap.insert(std::make_pair(SCEVCallbackVH(PN, this), SymbolicName));
if (StructType *STy = dyn_cast<StructType>(*GTI++)) {
// For a struct, add the member offset.
unsigned FieldNo = cast<ConstantInt>(Index)->getZExtValue();
- const SCEV *FieldOffset = getOffsetOfExpr(STy, FieldNo);
+ const SCEV *FieldOffset = getOffsetOfExpr(STy, IntPtrTy, FieldNo);
// Add the field offset to the running total offset.
TotalOffset = getAddExpr(TotalOffset, FieldOffset);
} else {
// For an array, add the element offset, explicitly scaled.
- const SCEV *ElementSize = getSizeOfExpr(*GTI);
+ const SCEV *ElementSize = getSizeOfExpr(*GTI, IntPtrTy);
const SCEV *IndexS = getSCEV(Index);
// Getelementptr indices are signed.
IndexS = getTruncateOrSignExtend(IndexS, IntPtrTy);
ConstantInt *Result = MulC->getValue();
- // Guard against huge trip counts.
- if (!Result || Result->getValue().getActiveBits() > 32)
+ // Guard against huge trip counts (this requires checking
+ // for zero to handle the case where the trip count == -1 and the
+ // addition wraps).
+ if (!Result || Result->getValue().getActiveBits() > 32 ||
+ Result->getValue().getActiveBits() == 0)
return 1;
return (unsigned)Result->getZExtValue();
if (!Visited.insert(I)) continue;
ValueExprMapType::iterator It =
- ValueExprMap.find(static_cast<Value *>(I));
+ ValueExprMap.find_as(static_cast<Value *>(I));
if (It != ValueExprMap.end()) {
const SCEV *Old = It->second;
Instruction *I = Worklist.pop_back_val();
if (!Visited.insert(I)) continue;
- ValueExprMapType::iterator It = ValueExprMap.find(static_cast<Value *>(I));
+ ValueExprMapType::iterator It =
+ ValueExprMap.find_as(static_cast<Value *>(I));
if (It != ValueExprMap.end()) {
forgetMemoizedResults(It->second);
ValueExprMap.erase(It);
I = Worklist.pop_back_val();
if (!Visited.insert(I)) continue;
- ValueExprMapType::iterator It = ValueExprMap.find(static_cast<Value *>(I));
+ ValueExprMapType::iterator It =
+ ValueExprMap.find_as(static_cast<Value *>(I));
if (It != ValueExprMap.end()) {
forgetMemoizedResults(It->second);
ValueExprMap.erase(It);
/// reason, return null.
static Constant *EvaluateExpression(Value *V, const Loop *L,
DenseMap<Instruction *, Constant *> &Vals,
- const TargetData *TD,
+ const DataLayout *TD,
const TargetLibraryInfo *TLI) {
// Convenient constant check, but redundant for recursive calls.
if (Constant *C = dyn_cast<Constant>(V)) return C;
SqrtTerm *= B;
SqrtTerm -= Four * (A * C);
+ if (SqrtTerm.isNegative()) {
+ // The loop is provably infinite.
+ const SCEV *CNC = SE.getCouldNotCompute();
+ return std::make_pair(CNC, CNC);
+ }
+
// Compute sqrt(B^2-4ac). This is guaranteed to be the nearest
// integer value or else APInt::sqrt() will assert.
APInt SqrtVal(SqrtTerm.sqrt());
// to 0, it must be counting down to equal 0. Consequently, N = Start / -Step.
// We have not yet seen any such cases.
const SCEVConstant *StepC = dyn_cast<SCEVConstant>(Step);
- if (StepC == 0)
+ if (StepC == 0 || StepC->getValue()->equalsInt(0))
return getCouldNotCompute();
// For positive steps (counting up until unsigned overflow):
if (!RA)
if (const SCEVAddExpr *AE = dyn_cast<SCEVAddExpr>(LHS))
if (const SCEVMulExpr *ME = dyn_cast<SCEVMulExpr>(AE->getOperand(0)))
- if (ME->getOperand(0)->isAllOnesValue()) {
+ if (AE->getNumOperands() == 2 && ME->getNumOperands() == 2 &&
+ ME->getOperand(0)->isAllOnesValue()) {
RHS = AE->getOperand(1);
LHS = ME->getOperand(1);
Changed = true;
bool ScalarEvolution::runOnFunction(Function &F) {
this->F = &F;
LI = &getAnalysis<LoopInfo>();
- TD = getAnalysisIfAvailable<TargetData>();
+ TD = getAnalysisIfAvailable<DataLayout>();
TLI = &getAnalysis<TargetLibraryInfo>();
DT = &getAnalysis<DominatorTree>();
return false;
return getBlockDisposition(S, BB) == ProperlyDominatesBlock;
}
-bool ScalarEvolution::hasOperand(const SCEV *S, const SCEV *Op) const {
- SmallVector<const SCEV *, 8> Worklist;
- Worklist.push_back(S);
- do {
- S = Worklist.pop_back_val();
+namespace {
+// Search for a SCEV expression node within an expression tree.
+// Implements SCEVTraversal::Visitor.
+struct SCEVSearch {
+ const SCEV *Node;
+ bool IsFound;
- switch (S->getSCEVType()) {
- case scConstant:
- break;
- case scTruncate:
- case scZeroExtend:
- case scSignExtend: {
- const SCEVCastExpr *Cast = cast<SCEVCastExpr>(S);
- const SCEV *CastOp = Cast->getOperand();
- if (Op == CastOp)
- return true;
- Worklist.push_back(CastOp);
- break;
- }
- case scAddRecExpr:
- case scAddExpr:
- case scMulExpr:
- case scUMaxExpr:
- case scSMaxExpr: {
- const SCEVNAryExpr *NAry = cast<SCEVNAryExpr>(S);
- for (SCEVNAryExpr::op_iterator I = NAry->op_begin(), E = NAry->op_end();
- I != E; ++I) {
- const SCEV *NAryOp = *I;
- if (NAryOp == Op)
- return true;
- Worklist.push_back(NAryOp);
- }
- break;
- }
- case scUDivExpr: {
- const SCEVUDivExpr *UDiv = cast<SCEVUDivExpr>(S);
- const SCEV *LHS = UDiv->getLHS(), *RHS = UDiv->getRHS();
- if (LHS == Op || RHS == Op)
- return true;
- Worklist.push_back(LHS);
- Worklist.push_back(RHS);
- break;
- }
- case scUnknown:
- break;
- case scCouldNotCompute:
- llvm_unreachable("Attempt to use a SCEVCouldNotCompute object!");
- default:
- llvm_unreachable("Unknown SCEV kind!");
- }
- } while (!Worklist.empty());
+ SCEVSearch(const SCEV *N): Node(N), IsFound(false) {}
- return false;
+ bool follow(const SCEV *S) {
+ IsFound |= (S == Node);
+ return !IsFound;
+ }
+ bool isDone() const { return IsFound; }
+};
+}
+
+bool ScalarEvolution::hasOperand(const SCEV *S, const SCEV *Op) const {
+ SCEVSearch Search(Op);
+ visitAll(S, Search);
+ return Search.IsFound;
}
void ScalarEvolution::forgetMemoizedResults(const SCEV *S) {
UnsignedRanges.erase(S);
SignedRanges.erase(S);
}
+
+typedef DenseMap<const Loop *, std::string> VerifyMap;
+/// getLoopBackedgeTakenCounts - Helper method for verifyAnalysis.
+static void
+getLoopBackedgeTakenCounts(Loop *L, VerifyMap &Map, ScalarEvolution &SE) {
+ for (Loop::reverse_iterator I = L->rbegin(), E = L->rend(); I != E; ++I) {
+ getLoopBackedgeTakenCounts(*I, Map, SE); // recurse.
+
+ std::string &S = Map[L];
+ if (S.empty()) {
+ raw_string_ostream OS(S);
+ SE.getBackedgeTakenCount(L)->print(OS);
+ }
+ }
+}
+
+void ScalarEvolution::verifyAnalysis() const {
+ if (!VerifySCEV)
+ return;
+
+ ScalarEvolution &SE = *const_cast<ScalarEvolution *>(this);
+
+ // Gather stringified backedge taken counts for all loops using SCEV's caches.
+ // FIXME: It would be much better to store actual values instead of strings,
+ // but SCEV pointers will change if we drop the caches.
+ VerifyMap BackedgeDumpsOld, BackedgeDumpsNew;
+ for (LoopInfo::reverse_iterator I = LI->rbegin(), E = LI->rend(); I != E; ++I)
+ getLoopBackedgeTakenCounts(*I, BackedgeDumpsOld, SE);
+
+ // Gather stringified backedge taken counts for all loops without using
+ // SCEV's caches.
+ SE.releaseMemory();
+ for (LoopInfo::reverse_iterator I = LI->rbegin(), E = LI->rend(); I != E; ++I)
+ getLoopBackedgeTakenCounts(*I, BackedgeDumpsNew, SE);
+
+ // Now compare whether they're the same with and without caches. This allows
+ // verifying that no pass changed the cache.
+ assert(BackedgeDumpsOld.size() == BackedgeDumpsNew.size() &&
+ "New loops suddenly appeared!");
+
+ for (VerifyMap::iterator OldI = BackedgeDumpsOld.begin(),
+ OldE = BackedgeDumpsOld.end(),
+ NewI = BackedgeDumpsNew.begin();
+ OldI != OldE; ++OldI, ++NewI) {
+ assert(OldI->first == NewI->first && "Loop order changed!");
+
+ // Compare the stringified SCEVs. We don't care if undef backedgetaken count
+ // changes.
+ // FIXME: We currently ignore SCEV changes towards CouldNotCompute. This
+ // means that a pass is buggy or SCEV has to learn a new pattern but is
+ // usually not harmful.
+ if (OldI->second != NewI->second &&
+ OldI->second.find("undef") == std::string::npos &&
+ NewI->second != "***COULDNOTCOMPUTE***") {
+ dbgs() << "SCEVValidator: SCEV for Loop '"
+ << OldI->first->getHeader()->getName()
+ << "' from '" << OldI->second << "' to '" << NewI->second << "'!";
+ std::abort();
+ }
+ }
+
+ // TODO: Verify more things.
+}