//
//===----------------------------------------------------------------------===//
-#define DEBUG_TYPE "indvars"
#include "llvm/Transforms/Scalar.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
-#include "llvm/Analysis/Dominators.h"
+#include "llvm/Analysis/GlobalsModRef.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/LoopPass.h"
#include "llvm/Analysis/ScalarEvolutionExpander.h"
+#include "llvm/Analysis/ScalarEvolutionAliasAnalysis.h"
+#include "llvm/Analysis/TargetLibraryInfo.h"
+#include "llvm/Analysis/TargetTransformInfo.h"
#include "llvm/IR/BasicBlock.h"
+#include "llvm/IR/CFG.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/Dominators.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/LLVMContext.h"
+#include "llvm/IR/PatternMatch.h"
#include "llvm/IR/Type.h"
-#include "llvm/Support/CFG.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
-#include "llvm/Target/TargetLibraryInfo.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
#include "llvm/Transforms/Utils/Local.h"
+#include "llvm/Transforms/Utils/LoopUtils.h"
#include "llvm/Transforms/Utils/SimplifyIndVar.h"
using namespace llvm;
+#define DEBUG_TYPE "indvars"
+
STATISTIC(NumWidened , "Number of indvars widened");
STATISTIC(NumReplaced , "Number of exit values replaced");
STATISTIC(NumLFTR , "Number of loop exit tests replaced");
"verify-indvars", cl::Hidden,
cl::desc("Verify the ScalarEvolution result after running indvars"));
+static cl::opt<bool> ReduceLiveIVs("liv-reduce", cl::Hidden,
+ cl::desc("Reduce live induction variables."));
+
+enum ReplaceExitVal { NeverRepl, OnlyCheapRepl, AlwaysRepl };
+
+static cl::opt<ReplaceExitVal> ReplaceExitValue(
+ "replexitval", cl::Hidden, cl::init(OnlyCheapRepl),
+ cl::desc("Choose the strategy to replace exit value in IndVarSimplify"),
+ cl::values(clEnumValN(NeverRepl, "never", "never replace exit value"),
+ clEnumValN(OnlyCheapRepl, "cheap",
+ "only replace exit value when the cost is cheap"),
+ clEnumValN(AlwaysRepl, "always",
+ "always replace exit value whenever possible"),
+ clEnumValEnd));
+
namespace {
- class IndVarSimplify : public LoopPass {
- LoopInfo *LI;
- ScalarEvolution *SE;
- DominatorTree *DT;
- DataLayout *TD;
- TargetLibraryInfo *TLI;
-
- SmallVector<WeakVH, 16> DeadInsts;
- bool Changed;
- public:
-
- static char ID; // Pass identification, replacement for typeid
- IndVarSimplify() : LoopPass(ID), LI(0), SE(0), DT(0), TD(0),
- Changed(false) {
- initializeIndVarSimplifyPass(*PassRegistry::getPassRegistry());
- }
+struct RewritePhi;
- virtual bool runOnLoop(Loop *L, LPPassManager &LPM);
-
- virtual void getAnalysisUsage(AnalysisUsage &AU) const {
- AU.addRequired<DominatorTree>();
- AU.addRequired<LoopInfo>();
- AU.addRequired<ScalarEvolution>();
- AU.addRequiredID(LoopSimplifyID);
- AU.addRequiredID(LCSSAID);
- AU.addPreserved<ScalarEvolution>();
- AU.addPreservedID(LoopSimplifyID);
- AU.addPreservedID(LCSSAID);
- AU.setPreservesCFG();
- }
+class IndVarSimplify : public LoopPass {
+ LoopInfo *LI;
+ ScalarEvolution *SE;
+ DominatorTree *DT;
+ TargetLibraryInfo *TLI;
+ const TargetTransformInfo *TTI;
- private:
- virtual void releaseMemory() {
- DeadInsts.clear();
- }
+ SmallVector<WeakVH, 16> DeadInsts;
+ bool Changed;
+public:
- bool isValidRewrite(Value *FromVal, Value *ToVal);
+ static char ID; // Pass identification, replacement for typeid
+ IndVarSimplify()
+ : LoopPass(ID), LI(nullptr), SE(nullptr), DT(nullptr), Changed(false) {
+ initializeIndVarSimplifyPass(*PassRegistry::getPassRegistry());
+ }
- void HandleFloatingPointIV(Loop *L, PHINode *PH);
- void RewriteNonIntegerIVs(Loop *L);
+ bool runOnLoop(Loop *L, LPPassManager &LPM) override;
+
+ void getAnalysisUsage(AnalysisUsage &AU) const override {
+ AU.addRequired<DominatorTreeWrapperPass>();
+ AU.addRequired<LoopInfoWrapperPass>();
+ AU.addRequired<ScalarEvolutionWrapperPass>();
+ AU.addRequiredID(LoopSimplifyID);
+ AU.addRequiredID(LCSSAID);
+ AU.addPreserved<GlobalsAAWrapperPass>();
+ AU.addPreserved<ScalarEvolutionWrapperPass>();
+ AU.addPreservedID(LoopSimplifyID);
+ AU.addPreservedID(LCSSAID);
+ AU.setPreservesCFG();
+ }
- void SimplifyAndExtend(Loop *L, SCEVExpander &Rewriter, LPPassManager &LPM);
+private:
+ void releaseMemory() override {
+ DeadInsts.clear();
+ }
- void RewriteLoopExitValues(Loop *L, SCEVExpander &Rewriter);
+ bool isValidRewrite(Value *FromVal, Value *ToVal);
- Value *LinearFunctionTestReplace(Loop *L, const SCEV *BackedgeTakenCount,
- PHINode *IndVar, SCEVExpander &Rewriter);
+ void handleFloatingPointIV(Loop *L, PHINode *PH);
+ void rewriteNonIntegerIVs(Loop *L);
- void SinkUnusedInvariants(Loop *L);
- };
+ void simplifyAndExtend(Loop *L, SCEVExpander &Rewriter, LoopInfo *LI);
+
+ bool canLoopBeDeleted(Loop *L, SmallVector<RewritePhi, 8> &RewritePhiSet);
+ void rewriteLoopExitValues(Loop *L, SCEVExpander &Rewriter);
+
+ Value *linearFunctionTestReplace(Loop *L, const SCEV *BackedgeTakenCount,
+ PHINode *IndVar, SCEVExpander &Rewriter);
+
+ void sinkUnusedInvariants(Loop *L);
+
+ Value *expandSCEVIfNeeded(SCEVExpander &Rewriter, const SCEV *S, Loop *L,
+ Instruction *InsertPt, Type *Ty);
+};
}
char IndVarSimplify::ID = 0;
INITIALIZE_PASS_BEGIN(IndVarSimplify, "indvars",
"Induction Variable Simplification", false, false)
-INITIALIZE_PASS_DEPENDENCY(DominatorTree)
-INITIALIZE_PASS_DEPENDENCY(LoopInfo)
-INITIALIZE_PASS_DEPENDENCY(ScalarEvolution)
+INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
+INITIALIZE_PASS_DEPENDENCY(LoopInfoWrapperPass)
+INITIALIZE_PASS_DEPENDENCY(ScalarEvolutionWrapperPass)
INITIALIZE_PASS_DEPENDENCY(LoopSimplify)
INITIALIZE_PASS_DEPENDENCY(LCSSA)
INITIALIZE_PASS_END(IndVarSimplify, "indvars",
return new IndVarSimplify();
}
-/// isValidRewrite - Return true if the SCEV expansion generated by the
-/// rewriter can replace the original value. SCEV guarantees that it
-/// produces the same value, but the way it is produced may be illegal IR.
-/// Ideally, this function will only be called for verification.
+/// Return true if the SCEV expansion generated by the rewriter can replace the
+/// original value. SCEV guarantees that it produces the same value, but the way
+/// it is produced may be illegal IR. Ideally, this function will only be
+/// called for verification.
bool IndVarSimplify::isValidRewrite(Value *FromVal, Value *ToVal) {
// If an SCEV expression subsumed multiple pointers, its expansion could
// reassociate the GEP changing the base pointer. This is illegal because the
// because it understands lcssa phis while SCEV does not.
Value *FromPtr = FromVal;
Value *ToPtr = ToVal;
- if (GEPOperator *GEP = dyn_cast<GEPOperator>(FromVal)) {
+ if (auto *GEP = dyn_cast<GEPOperator>(FromVal)) {
FromPtr = GEP->getPointerOperand();
}
- if (GEPOperator *GEP = dyn_cast<GEPOperator>(ToVal)) {
+ if (auto *GEP = dyn_cast<GEPOperator>(ToVal)) {
ToPtr = GEP->getPointerOperand();
}
if (FromPtr != FromVal || ToPtr != ToVal) {
/// loop. For PHI nodes, there may be multiple uses, so compute the nearest
/// common dominator for the incoming blocks.
static Instruction *getInsertPointForUses(Instruction *User, Value *Def,
- DominatorTree *DT) {
+ DominatorTree *DT, LoopInfo *LI) {
PHINode *PHI = dyn_cast<PHINode>(User);
if (!PHI)
return User;
- Instruction *InsertPt = 0;
+ Instruction *InsertPt = nullptr;
for (unsigned i = 0, e = PHI->getNumIncomingValues(); i != e; ++i) {
if (PHI->getIncomingValue(i) != Def)
continue;
InsertPt = InsertBB->getTerminator();
}
assert(InsertPt && "Missing phi operand");
- assert((!isa<Instruction>(Def) ||
- DT->dominates(cast<Instruction>(Def), InsertPt)) &&
- "def does not dominate all uses");
- return InsertPt;
+
+ auto *DefI = dyn_cast<Instruction>(Def);
+ if (!DefI)
+ return InsertPt;
+
+ assert(DT->dominates(DefI, InsertPt) && "def does not dominate all uses");
+
+ auto *L = LI->getLoopFor(DefI->getParent());
+ assert(!L || L->contains(LI->getLoopFor(InsertPt->getParent())));
+
+ for (auto *DTN = (*DT)[InsertPt->getParent()]; DTN; DTN = DTN->getIDom())
+ if (LI->getLoopFor(DTN->getBlock()) == L)
+ return DTN->getBlock()->getTerminator();
+
+ llvm_unreachable("DefI dominates InsertPt!");
}
//===----------------------------------------------------------------------===//
-// RewriteNonIntegerIVs and helpers. Prefer integer IVs.
+// rewriteNonIntegerIVs and helpers. Prefer integer IVs.
//===----------------------------------------------------------------------===//
-/// ConvertToSInt - Convert APF to an integer, if possible.
+/// Convert APF to an integer, if possible.
static bool ConvertToSInt(const APFloat &APF, int64_t &IntVal) {
bool isExact = false;
// See if we can convert this to an int64_t
return true;
}
-/// HandleFloatingPointIV - If the loop has floating induction variable
-/// then insert corresponding integer induction variable if possible.
+/// If the loop has floating induction variable then insert corresponding
+/// integer induction variable if possible.
/// For example,
/// for(double i = 0; i < 10000; ++i)
/// bar(i)
/// for(int i = 0; i < 10000; ++i)
/// bar((double)i);
///
-void IndVarSimplify::HandleFloatingPointIV(Loop *L, PHINode *PN) {
+void IndVarSimplify::handleFloatingPointIV(Loop *L, PHINode *PN) {
unsigned IncomingEdge = L->contains(PN->getIncomingBlock(0));
unsigned BackEdge = IncomingEdge^1;
// Check incoming value.
- ConstantFP *InitValueVal =
- dyn_cast<ConstantFP>(PN->getIncomingValue(IncomingEdge));
+ auto *InitValueVal = dyn_cast<ConstantFP>(PN->getIncomingValue(IncomingEdge));
int64_t InitValue;
if (!InitValueVal || !ConvertToSInt(InitValueVal->getValueAPF(), InitValue))
// Check IV increment. Reject this PN if increment operation is not
// an add or increment value can not be represented by an integer.
- BinaryOperator *Incr =
- dyn_cast<BinaryOperator>(PN->getIncomingValue(BackEdge));
- if (Incr == 0 || Incr->getOpcode() != Instruction::FAdd) return;
+ auto *Incr = dyn_cast<BinaryOperator>(PN->getIncomingValue(BackEdge));
+ if (Incr == nullptr || Incr->getOpcode() != Instruction::FAdd) return;
// If this is not an add of the PHI with a constantfp, or if the constant fp
// is not an integer, bail out.
ConstantFP *IncValueVal = dyn_cast<ConstantFP>(Incr->getOperand(1));
int64_t IncValue;
- if (IncValueVal == 0 || Incr->getOperand(0) != PN ||
+ if (IncValueVal == nullptr || Incr->getOperand(0) != PN ||
!ConvertToSInt(IncValueVal->getValueAPF(), IncValue))
return;
// Check Incr uses. One user is PN and the other user is an exit condition
// used by the conditional terminator.
- Value::use_iterator IncrUse = Incr->use_begin();
+ Value::user_iterator IncrUse = Incr->user_begin();
Instruction *U1 = cast<Instruction>(*IncrUse++);
- if (IncrUse == Incr->use_end()) return;
+ if (IncrUse == Incr->user_end()) return;
Instruction *U2 = cast<Instruction>(*IncrUse++);
- if (IncrUse != Incr->use_end()) return;
+ if (IncrUse != Incr->user_end()) return;
// Find exit condition, which is an fcmp. If it doesn't exist, or if it isn't
// only used by a branch, we can't transform it.
FCmpInst *Compare = dyn_cast<FCmpInst>(U1);
if (!Compare)
Compare = dyn_cast<FCmpInst>(U2);
- if (Compare == 0 || !Compare->hasOneUse() ||
- !isa<BranchInst>(Compare->use_back()))
+ if (!Compare || !Compare->hasOneUse() ||
+ !isa<BranchInst>(Compare->user_back()))
return;
- BranchInst *TheBr = cast<BranchInst>(Compare->use_back());
+ BranchInst *TheBr = cast<BranchInst>(Compare->user_back());
// We need to verify that the branch actually controls the iteration count
// of the loop. If not, the new IV can overflow and no one will notice.
// transform it.
ConstantFP *ExitValueVal = dyn_cast<ConstantFP>(Compare->getOperand(1));
int64_t ExitValue;
- if (ExitValueVal == 0 ||
+ if (ExitValueVal == nullptr ||
!ConvertToSInt(ExitValueVal->getValueAPF(), ExitValue))
return;
// platforms.
if (WeakPH) {
Value *Conv = new SIToFPInst(NewPHI, PN->getType(), "indvar.conv",
- PN->getParent()->getFirstInsertionPt());
+ &*PN->getParent()->getFirstInsertionPt());
PN->replaceAllUsesWith(Conv);
RecursivelyDeleteTriviallyDeadInstructions(PN, TLI);
}
Changed = true;
}
-void IndVarSimplify::RewriteNonIntegerIVs(Loop *L) {
+void IndVarSimplify::rewriteNonIntegerIVs(Loop *L) {
// First step. Check to see if there are any floating-point recurrences.
// If there are, change them into integer recurrences, permitting analysis by
// the SCEV routines.
for (unsigned i = 0, e = PHIs.size(); i != e; ++i)
if (PHINode *PN = dyn_cast_or_null<PHINode>(&*PHIs[i]))
- HandleFloatingPointIV(L, PN);
+ handleFloatingPointIV(L, PN);
// If the loop previously had floating-point IV, ScalarEvolution
// may not have been able to compute a trip count. Now that we've done some
SE->forgetLoop(L);
}
+namespace {
+// Collect information about PHI nodes which can be transformed in
+// rewriteLoopExitValues.
+struct RewritePhi {
+ PHINode *PN;
+ unsigned Ith; // Ith incoming value.
+ Value *Val; // Exit value after expansion.
+ bool HighCost; // High Cost when expansion.
+ bool SafePhi; // LCSSASafePhiForRAUW.
+
+ RewritePhi(PHINode *P, unsigned I, Value *V, bool H, bool S)
+ : PN(P), Ith(I), Val(V), HighCost(H), SafePhi(S) {}
+};
+}
+
+Value *IndVarSimplify::expandSCEVIfNeeded(SCEVExpander &Rewriter, const SCEV *S,
+ Loop *L, Instruction *InsertPt,
+ Type *ResultTy) {
+ // Before expanding S into an expensive LLVM expression, see if we can use an
+ // already existing value as the expansion for S.
+ if (Value *ExistingValue = Rewriter.findExistingExpansion(S, InsertPt, L))
+ if (ExistingValue->getType() == ResultTy)
+ return ExistingValue;
+
+ // We didn't find anything, fall back to using SCEVExpander.
+ return Rewriter.expandCodeFor(S, ResultTy, InsertPt);
+}
+
//===----------------------------------------------------------------------===//
-// RewriteLoopExitValues - Optimize IV users outside the loop.
+// rewriteLoopExitValues - Optimize IV users outside the loop.
// As a side effect, reduces the amount of IV processing within the loop.
//===----------------------------------------------------------------------===//
-/// RewriteLoopExitValues - Check to see if this loop has a computable
-/// loop-invariant execution count. If so, this means that we can compute the
-/// final value of any expressions that are recurrent in the loop, and
-/// substitute the exit values from the loop into any instructions outside of
-/// the loop that use the final values of the current expressions.
+/// Check to see if this loop has a computable loop-invariant execution count.
+/// If so, this means that we can compute the final value of any expressions
+/// that are recurrent in the loop, and substitute the exit values from the loop
+/// into any instructions outside of the loop that use the final values of the
+/// current expressions.
///
/// This is mostly redundant with the regular IndVarSimplify activities that
/// happen later, except that it's more powerful in some cases, because it's
/// able to brute-force evaluate arbitrary instructions as long as they have
/// constant operands at the beginning of the loop.
-void IndVarSimplify::RewriteLoopExitValues(Loop *L, SCEVExpander &Rewriter) {
- // Verify the input to the pass in already in LCSSA form.
- assert(L->isLCSSAForm(*DT));
+void IndVarSimplify::rewriteLoopExitValues(Loop *L, SCEVExpander &Rewriter) {
+ // Check a pre-condition.
+ assert(L->isRecursivelyLCSSAForm(*DT) && "Indvars did not preserve LCSSA!");
SmallVector<BasicBlock*, 8> ExitBlocks;
L->getUniqueExitBlocks(ExitBlocks);
+ SmallVector<RewritePhi, 8> RewritePhiSet;
// Find all values that are computed inside the loop, but used outside of it.
// Because of LCSSA, these values will only occur in LCSSA PHI Nodes. Scan
// the exit blocks of the loop to find them.
unsigned NumPreds = PN->getNumIncomingValues();
+ // We would like to be able to RAUW single-incoming value PHI nodes. We
+ // have to be certain this is safe even when this is an LCSSA PHI node.
+ // While the computed exit value is no longer varying in *this* loop, the
+ // exit block may be an exit block for an outer containing loop as well,
+ // the exit value may be varying in the outer loop, and thus it may still
+ // require an LCSSA PHI node. The safe case is when this is
+ // single-predecessor PHI node (LCSSA) and the exit block containing it is
+ // part of the enclosing loop, or this is the outer most loop of the nest.
+ // In either case the exit value could (at most) be varying in the same
+ // loop body as the phi node itself. Thus if it is in turn used outside of
+ // an enclosing loop it will only be via a separate LCSSA node.
+ bool LCSSASafePhiForRAUW =
+ NumPreds == 1 &&
+ (!L->getParentLoop() || L->getParentLoop() == LI->getLoopFor(ExitBB));
+
// Iterate over all of the PHI nodes.
BasicBlock::iterator BBI = ExitBB->begin();
while ((PN = dyn_cast<PHINode>(BBI++))) {
unsigned NumHardInternalUses = 0;
unsigned NumSoftExternalUses = 0;
unsigned NumUses = 0;
- for (Value::use_iterator IB=Inst->use_begin(), IE=Inst->use_end();
- IB!=IE && NumUses<=6 ; ++IB) {
+ for (auto IB = Inst->user_begin(), IE = Inst->user_end();
+ IB != IE && NumUses <= 6; ++IB) {
Instruction *UseInstr = cast<Instruction>(*IB);
unsigned Opc = UseInstr->getOpcode();
NumUses++;
// Do not count the Phi as a use. LCSSA may have inserted
// plenty of trivial ones.
NumUses--;
- for (Value::use_iterator PB=UseInstr->use_begin(),
- PE=UseInstr->use_end();
- PB!=PE && NumUses<=6 ; ++PB, ++NumUses) {
+ for (auto PB = UseInstr->user_begin(),
+ PE = UseInstr->user_end();
+ PB != PE && NumUses <= 6; ++PB, ++NumUses) {
unsigned PhiOpc = cast<Instruction>(*PB)->getOpcode();
if (PhiOpc != Instruction::Call && PhiOpc != Instruction::Ret)
NumSoftExternalUses++;
continue;
}
- Value *ExitVal = Rewriter.expandCodeFor(ExitValue, PN->getType(), Inst);
+ bool HighCost = Rewriter.isHighCostExpansion(ExitValue, L, Inst);
+ Value *ExitVal =
+ expandSCEVIfNeeded(Rewriter, ExitValue, L, Inst, PN->getType());
DEBUG(dbgs() << "INDVARS: RLEV: AfterLoopVal = " << *ExitVal << '\n'
<< " LoopVal = " << *Inst << "\n");
DeadInsts.push_back(ExitVal);
continue;
}
- Changed = true;
- ++NumReplaced;
- PN->setIncomingValue(i, ExitVal);
+ // Collect all the candidate PHINodes to be rewritten.
+ RewritePhiSet.push_back(
+ RewritePhi(PN, i, ExitVal, HighCost, LCSSASafePhiForRAUW));
+ }
+ }
+ }
- // If this instruction is dead now, delete it. Don't do it now to avoid
- // invalidating iterators.
- if (isInstructionTriviallyDead(Inst, TLI))
- DeadInsts.push_back(Inst);
+ bool LoopCanBeDel = canLoopBeDeleted(L, RewritePhiSet);
- if (NumPreds == 1) {
- // Completely replace a single-pred PHI. This is safe, because the
- // NewVal won't be variant in the loop, so we don't need an LCSSA phi
- // node anymore.
- PN->replaceAllUsesWith(ExitVal);
- PN->eraseFromParent();
- }
- }
- if (NumPreds != 1) {
- // Clone the PHI and delete the original one. This lets IVUsers and
- // any other maps purge the original user from their records.
- PHINode *NewPN = cast<PHINode>(PN->clone());
- NewPN->takeName(PN);
- NewPN->insertBefore(PN);
- PN->replaceAllUsesWith(NewPN);
- PN->eraseFromParent();
- }
+ // Transformation.
+ for (const RewritePhi &Phi : RewritePhiSet) {
+ PHINode *PN = Phi.PN;
+ Value *ExitVal = Phi.Val;
+
+ // Only do the rewrite when the ExitValue can be expanded cheaply.
+ // If LoopCanBeDel is true, rewrite exit value aggressively.
+ if (ReplaceExitValue == OnlyCheapRepl && !LoopCanBeDel && Phi.HighCost) {
+ DeadInsts.push_back(ExitVal);
+ continue;
+ }
+
+ Changed = true;
+ ++NumReplaced;
+ Instruction *Inst = cast<Instruction>(PN->getIncomingValue(Phi.Ith));
+ PN->setIncomingValue(Phi.Ith, ExitVal);
+
+ // If this instruction is dead now, delete it. Don't do it now to avoid
+ // invalidating iterators.
+ if (isInstructionTriviallyDead(Inst, TLI))
+ DeadInsts.push_back(Inst);
+
+ // If we determined that this PHI is safe to replace even if an LCSSA
+ // PHI, do so.
+ if (Phi.SafePhi) {
+ PN->replaceAllUsesWith(ExitVal);
+ PN->eraseFromParent();
}
}
Rewriter.clearInsertPoint();
}
-//===----------------------------------------------------------------------===//
-// IV Widening - Extend the width of an IV to cover its widest uses.
-//===----------------------------------------------------------------------===//
+/// Check whether it is possible to delete the loop after rewriting exit
+/// value. If it is possible, ignore ReplaceExitValue and do rewriting
+/// aggressively.
+bool IndVarSimplify::canLoopBeDeleted(
+ Loop *L, SmallVector<RewritePhi, 8> &RewritePhiSet) {
-namespace {
- // Collect information about induction variables that are used by sign/zero
- // extend operations. This information is recorded by CollectExtend and
- // provides the input to WidenIV.
- struct WideIVInfo {
- PHINode *NarrowIV;
- Type *WidestNativeType; // Widest integer type created [sz]ext
- bool IsSigned; // Was an sext user seen before a zext?
-
- WideIVInfo() : NarrowIV(0), WidestNativeType(0), IsSigned(false) {}
- };
+ BasicBlock *Preheader = L->getLoopPreheader();
+ // If there is no preheader, the loop will not be deleted.
+ if (!Preheader)
+ return false;
- class WideIVVisitor : public IVVisitor {
- ScalarEvolution *SE;
- const DataLayout *TD;
+ // In LoopDeletion pass Loop can be deleted when ExitingBlocks.size() > 1.
+ // We obviate multiple ExitingBlocks case for simplicity.
+ // TODO: If we see testcase with multiple ExitingBlocks can be deleted
+ // after exit value rewriting, we can enhance the logic here.
+ SmallVector<BasicBlock *, 4> ExitingBlocks;
+ L->getExitingBlocks(ExitingBlocks);
+ SmallVector<BasicBlock *, 8> ExitBlocks;
+ L->getUniqueExitBlocks(ExitBlocks);
+ if (ExitBlocks.size() > 1 || ExitingBlocks.size() > 1)
+ return false;
- public:
- WideIVInfo WI;
+ BasicBlock *ExitBlock = ExitBlocks[0];
+ BasicBlock::iterator BI = ExitBlock->begin();
+ while (PHINode *P = dyn_cast<PHINode>(BI)) {
+ Value *Incoming = P->getIncomingValueForBlock(ExitingBlocks[0]);
+
+ // If the Incoming value of P is found in RewritePhiSet, we know it
+ // could be rewritten to use a loop invariant value in transformation
+ // phase later. Skip it in the loop invariant check below.
+ bool found = false;
+ for (const RewritePhi &Phi : RewritePhiSet) {
+ unsigned i = Phi.Ith;
+ if (Phi.PN == P && (Phi.PN)->getIncomingValue(i) == Incoming) {
+ found = true;
+ break;
+ }
+ }
- WideIVVisitor(PHINode *NarrowIV, ScalarEvolution *SCEV,
- const DataLayout *TData) :
- SE(SCEV), TD(TData) { WI.NarrowIV = NarrowIV; }
+ Instruction *I;
+ if (!found && (I = dyn_cast<Instruction>(Incoming)))
+ if (!L->hasLoopInvariantOperands(I))
+ return false;
- // Implement the interface used by simplifyUsersOfIV.
- virtual void visitCast(CastInst *Cast);
- };
+ ++BI;
+ }
+
+ for (auto *BB : L->blocks())
+ if (any_of(*BB, [](Instruction &I) { return I.mayHaveSideEffects(); }))
+ return false;
+
+ return true;
}
-/// visitCast - Update information about the induction variable that is
-/// extended by this sign or zero extend operation. This is used to determine
-/// the final width of the IV before actually widening it.
-void WideIVVisitor::visitCast(CastInst *Cast) {
+//===----------------------------------------------------------------------===//
+// IV Widening - Extend the width of an IV to cover its widest uses.
+//===----------------------------------------------------------------------===//
+
+namespace {
+// Collect information about induction variables that are used by sign/zero
+// extend operations. This information is recorded by CollectExtend and provides
+// the input to WidenIV.
+struct WideIVInfo {
+ PHINode *NarrowIV = nullptr;
+ Type *WidestNativeType = nullptr; // Widest integer type created [sz]ext
+ bool IsSigned = false; // Was a sext user seen before a zext?
+};
+}
+
+/// Update information about the induction variable that is extended by this
+/// sign or zero extend operation. This is used to determine the final width of
+/// the IV before actually widening it.
+static void visitIVCast(CastInst *Cast, WideIVInfo &WI, ScalarEvolution *SE,
+ const TargetTransformInfo *TTI) {
bool IsSigned = Cast->getOpcode() == Instruction::SExt;
if (!IsSigned && Cast->getOpcode() != Instruction::ZExt)
return;
Type *Ty = Cast->getType();
uint64_t Width = SE->getTypeSizeInBits(Ty);
- if (TD && !TD->isLegalInteger(Width))
+ if (!Cast->getModule()->getDataLayout().isLegalInteger(Width))
return;
+ // Cast is either an sext or zext up to this point.
+ // We should not widen an indvar if arithmetics on the wider indvar are more
+ // expensive than those on the narrower indvar. We check only the cost of ADD
+ // because at least an ADD is required to increment the induction variable. We
+ // could compute more comprehensively the cost of all instructions on the
+ // induction variable when necessary.
+ if (TTI &&
+ TTI->getArithmeticInstrCost(Instruction::Add, Ty) >
+ TTI->getArithmeticInstrCost(Instruction::Add,
+ Cast->getOperand(0)->getType())) {
+ return;
+ }
+
if (!WI.WidestNativeType) {
WI.WidestNativeType = SE->getEffectiveSCEVType(Ty);
WI.IsSigned = IsSigned;
namespace {
-/// NarrowIVDefUse - Record a link in the Narrow IV def-use chain along with the
-/// WideIV that computes the same value as the Narrow IV def. This avoids
-/// caching Use* pointers.
+/// Record a link in the Narrow IV def-use chain along with the WideIV that
+/// computes the same value as the Narrow IV def. This avoids caching Use*
+/// pointers.
struct NarrowIVDefUse {
- Instruction *NarrowDef;
- Instruction *NarrowUse;
- Instruction *WideDef;
-
- NarrowIVDefUse(): NarrowDef(0), NarrowUse(0), WideDef(0) {}
-
- NarrowIVDefUse(Instruction *ND, Instruction *NU, Instruction *WD):
- NarrowDef(ND), NarrowUse(NU), WideDef(WD) {}
+ Instruction *NarrowDef = nullptr;
+ Instruction *NarrowUse = nullptr;
+ Instruction *WideDef = nullptr;
+
+ // True if the narrow def is never negative. Tracking this information lets
+ // us use a sign extension instead of a zero extension or vice versa, when
+ // profitable and legal.
+ bool NeverNegative = false;
+
+ NarrowIVDefUse(Instruction *ND, Instruction *NU, Instruction *WD,
+ bool NeverNegative)
+ : NarrowDef(ND), NarrowUse(NU), WideDef(WD),
+ NeverNegative(NeverNegative) {}
};
-/// WidenIV - The goal of this transform is to remove sign and zero extends
-/// without creating any new induction variables. To do this, it creates a new
-/// phi of the wider type and redirects all users, either removing extends or
-/// inserting truncs whenever we stop propagating the type.
+/// The goal of this transform is to remove sign and zero extends without
+/// creating any new induction variables. To do this, it creates a new phi of
+/// the wider type and redirects all users, either removing extends or inserting
+/// truncs whenever we stop propagating the type.
///
class WidenIV {
// Parameters
L(LI->getLoopFor(OrigPhi->getParent())),
SE(SEv),
DT(DTree),
- WidePhi(0),
- WideInc(0),
- WideIncExpr(0),
+ WidePhi(nullptr),
+ WideInc(nullptr),
+ WideIncExpr(nullptr),
DeadInsts(DI) {
assert(L->getHeader() == OrigPhi->getParent() && "Phi must be an IV");
}
- PHINode *CreateWideIV(SCEVExpander &Rewriter);
+ PHINode *createWideIV(SCEVExpander &Rewriter);
protected:
- Value *getExtend(Value *NarrowOper, Type *WideType, bool IsSigned,
- Instruction *Use);
+ Value *createExtendInst(Value *NarrowOper, Type *WideType, bool IsSigned,
+ Instruction *Use);
- Instruction *CloneIVUser(NarrowIVDefUse DU);
+ Instruction *cloneIVUser(NarrowIVDefUse DU, const SCEVAddRecExpr *WideAR);
+ Instruction *cloneArithmeticIVUser(NarrowIVDefUse DU,
+ const SCEVAddRecExpr *WideAR);
+ Instruction *cloneBitwiseIVUser(NarrowIVDefUse DU);
- const SCEVAddRecExpr *GetWideRecurrence(Instruction *NarrowUse);
+ const SCEVAddRecExpr *getWideRecurrence(Instruction *NarrowUse);
- const SCEVAddRecExpr* GetExtendedOperandRecurrence(NarrowIVDefUse DU);
+ const SCEVAddRecExpr* getExtendedOperandRecurrence(NarrowIVDefUse DU);
- Instruction *WidenIVUse(NarrowIVDefUse DU, SCEVExpander &Rewriter);
+ const SCEV *getSCEVByOpCode(const SCEV *LHS, const SCEV *RHS,
+ unsigned OpCode) const;
+
+ Instruction *widenIVUse(NarrowIVDefUse DU, SCEVExpander &Rewriter);
+
+ bool widenLoopCompare(NarrowIVDefUse DU);
void pushNarrowIVUsers(Instruction *NarrowDef, Instruction *WideDef);
};
} // anonymous namespace
-/// isLoopInvariant - Perform a quick domtree based check for loop invariance
-/// assuming that V is used within the loop. LoopInfo::isLoopInvariant() seems
-/// gratuitous for this purpose.
+/// Perform a quick domtree based check for loop invariance assuming that V is
+/// used within the loop. LoopInfo::isLoopInvariant() seems gratuitous for this
+/// purpose.
static bool isLoopInvariant(Value *V, const Loop *L, const DominatorTree *DT) {
Instruction *Inst = dyn_cast<Instruction>(V);
if (!Inst)
return DT->properlyDominates(Inst->getParent(), L->getHeader());
}
-Value *WidenIV::getExtend(Value *NarrowOper, Type *WideType, bool IsSigned,
- Instruction *Use) {
+Value *WidenIV::createExtendInst(Value *NarrowOper, Type *WideType,
+ bool IsSigned, Instruction *Use) {
// Set the debug location and conservative insertion point.
IRBuilder<> Builder(Use);
// Hoist the insertion point into loop preheaders as far as possible.
Builder.CreateZExt(NarrowOper, WideType);
}
-/// CloneIVUser - Instantiate a wide operation to replace a narrow
-/// operation. This only needs to handle operations that can evaluation to
-/// SCEVAddRec. It can safely return 0 for any operation we decide not to clone.
-Instruction *WidenIV::CloneIVUser(NarrowIVDefUse DU) {
+/// Instantiate a wide operation to replace a narrow operation. This only needs
+/// to handle operations that can evaluation to SCEVAddRec. It can safely return
+/// 0 for any operation we decide not to clone.
+Instruction *WidenIV::cloneIVUser(NarrowIVDefUse DU,
+ const SCEVAddRecExpr *WideAR) {
unsigned Opcode = DU.NarrowUse->getOpcode();
switch (Opcode) {
default:
- return 0;
+ return nullptr;
case Instruction::Add:
case Instruction::Mul:
case Instruction::UDiv:
case Instruction::Sub:
+ return cloneArithmeticIVUser(DU, WideAR);
+
case Instruction::And:
case Instruction::Or:
case Instruction::Xor:
case Instruction::Shl:
case Instruction::LShr:
case Instruction::AShr:
- DEBUG(dbgs() << "Cloning IVUser: " << *DU.NarrowUse << "\n");
-
- // Replace NarrowDef operands with WideDef. Otherwise, we don't know
- // anything about the narrow operand yet so must insert a [sz]ext. It is
- // probably loop invariant and will be folded or hoisted. If it actually
- // comes from a widened IV, it should be removed during a future call to
- // WidenIVUse.
- Value *LHS = (DU.NarrowUse->getOperand(0) == DU.NarrowDef) ? DU.WideDef :
- getExtend(DU.NarrowUse->getOperand(0), WideType, IsSigned, DU.NarrowUse);
- Value *RHS = (DU.NarrowUse->getOperand(1) == DU.NarrowDef) ? DU.WideDef :
- getExtend(DU.NarrowUse->getOperand(1), WideType, IsSigned, DU.NarrowUse);
-
- BinaryOperator *NarrowBO = cast<BinaryOperator>(DU.NarrowUse);
- BinaryOperator *WideBO = BinaryOperator::Create(NarrowBO->getOpcode(),
- LHS, RHS,
- NarrowBO->getName());
- IRBuilder<> Builder(DU.NarrowUse);
- Builder.Insert(WideBO);
- if (const OverflowingBinaryOperator *OBO =
- dyn_cast<OverflowingBinaryOperator>(NarrowBO)) {
- if (OBO->hasNoUnsignedWrap()) WideBO->setHasNoUnsignedWrap();
- if (OBO->hasNoSignedWrap()) WideBO->setHasNoSignedWrap();
+ return cloneBitwiseIVUser(DU);
+ }
+}
+
+Instruction *WidenIV::cloneBitwiseIVUser(NarrowIVDefUse DU) {
+ Instruction *NarrowUse = DU.NarrowUse;
+ Instruction *NarrowDef = DU.NarrowDef;
+ Instruction *WideDef = DU.WideDef;
+
+ DEBUG(dbgs() << "Cloning bitwise IVUser: " << *NarrowUse << "\n");
+
+ // Replace NarrowDef operands with WideDef. Otherwise, we don't know anything
+ // about the narrow operand yet so must insert a [sz]ext. It is probably loop
+ // invariant and will be folded or hoisted. If it actually comes from a
+ // widened IV, it should be removed during a future call to widenIVUse.
+ Value *LHS = (NarrowUse->getOperand(0) == NarrowDef)
+ ? WideDef
+ : createExtendInst(NarrowUse->getOperand(0), WideType,
+ IsSigned, NarrowUse);
+ Value *RHS = (NarrowUse->getOperand(1) == NarrowDef)
+ ? WideDef
+ : createExtendInst(NarrowUse->getOperand(1), WideType,
+ IsSigned, NarrowUse);
+
+ auto *NarrowBO = cast<BinaryOperator>(NarrowUse);
+ auto *WideBO = BinaryOperator::Create(NarrowBO->getOpcode(), LHS, RHS,
+ NarrowBO->getName());
+ IRBuilder<> Builder(NarrowUse);
+ Builder.Insert(WideBO);
+ WideBO->copyIRFlags(NarrowBO);
+ return WideBO;
+}
+
+Instruction *WidenIV::cloneArithmeticIVUser(NarrowIVDefUse DU,
+ const SCEVAddRecExpr *WideAR) {
+ Instruction *NarrowUse = DU.NarrowUse;
+ Instruction *NarrowDef = DU.NarrowDef;
+ Instruction *WideDef = DU.WideDef;
+
+ DEBUG(dbgs() << "Cloning arithmetic IVUser: " << *NarrowUse << "\n");
+
+ unsigned IVOpIdx = (NarrowUse->getOperand(0) == NarrowDef) ? 0 : 1;
+
+ // We're trying to find X such that
+ //
+ // Widen(NarrowDef `op` NonIVNarrowDef) == WideAR == WideDef `op.wide` X
+ //
+ // We guess two solutions to X, sext(NonIVNarrowDef) and zext(NonIVNarrowDef),
+ // and check using SCEV if any of them are correct.
+
+ // Returns true if extending NonIVNarrowDef according to `SignExt` is a
+ // correct solution to X.
+ auto GuessNonIVOperand = [&](bool SignExt) {
+ const SCEV *WideLHS;
+ const SCEV *WideRHS;
+
+ auto GetExtend = [this, SignExt](const SCEV *S, Type *Ty) {
+ if (SignExt)
+ return SE->getSignExtendExpr(S, Ty);
+ return SE->getZeroExtendExpr(S, Ty);
+ };
+
+ if (IVOpIdx == 0) {
+ WideLHS = SE->getSCEV(WideDef);
+ const SCEV *NarrowRHS = SE->getSCEV(NarrowUse->getOperand(1));
+ WideRHS = GetExtend(NarrowRHS, WideType);
+ } else {
+ const SCEV *NarrowLHS = SE->getSCEV(NarrowUse->getOperand(0));
+ WideLHS = GetExtend(NarrowLHS, WideType);
+ WideRHS = SE->getSCEV(WideDef);
}
- return WideBO;
+
+ // WideUse is "WideDef `op.wide` X" as described in the comment.
+ const SCEV *WideUse = nullptr;
+
+ switch (NarrowUse->getOpcode()) {
+ default:
+ llvm_unreachable("No other possibility!");
+
+ case Instruction::Add:
+ WideUse = SE->getAddExpr(WideLHS, WideRHS);
+ break;
+
+ case Instruction::Mul:
+ WideUse = SE->getMulExpr(WideLHS, WideRHS);
+ break;
+
+ case Instruction::UDiv:
+ WideUse = SE->getUDivExpr(WideLHS, WideRHS);
+ break;
+
+ case Instruction::Sub:
+ WideUse = SE->getMinusSCEV(WideLHS, WideRHS);
+ break;
+ }
+
+ return WideUse == WideAR;
+ };
+
+ bool SignExtend = IsSigned;
+ if (!GuessNonIVOperand(SignExtend)) {
+ SignExtend = !SignExtend;
+ if (!GuessNonIVOperand(SignExtend))
+ return nullptr;
}
+
+ Value *LHS = (NarrowUse->getOperand(0) == NarrowDef)
+ ? WideDef
+ : createExtendInst(NarrowUse->getOperand(0), WideType,
+ SignExtend, NarrowUse);
+ Value *RHS = (NarrowUse->getOperand(1) == NarrowDef)
+ ? WideDef
+ : createExtendInst(NarrowUse->getOperand(1), WideType,
+ SignExtend, NarrowUse);
+
+ auto *NarrowBO = cast<BinaryOperator>(NarrowUse);
+ auto *WideBO = BinaryOperator::Create(NarrowBO->getOpcode(), LHS, RHS,
+ NarrowBO->getName());
+
+ IRBuilder<> Builder(NarrowUse);
+ Builder.Insert(WideBO);
+ WideBO->copyIRFlags(NarrowBO);
+ return WideBO;
+}
+
+const SCEV *WidenIV::getSCEVByOpCode(const SCEV *LHS, const SCEV *RHS,
+ unsigned OpCode) const {
+ if (OpCode == Instruction::Add)
+ return SE->getAddExpr(LHS, RHS);
+ if (OpCode == Instruction::Sub)
+ return SE->getMinusSCEV(LHS, RHS);
+ if (OpCode == Instruction::Mul)
+ return SE->getMulExpr(LHS, RHS);
+
+ llvm_unreachable("Unsupported opcode.");
}
/// No-wrap operations can transfer sign extension of their result to their
/// operands. Generate the SCEV value for the widened operation without
/// actually modifying the IR yet. If the expression after extending the
/// operands is an AddRec for this loop, return it.
-const SCEVAddRecExpr* WidenIV::GetExtendedOperandRecurrence(NarrowIVDefUse DU) {
+const SCEVAddRecExpr* WidenIV::getExtendedOperandRecurrence(NarrowIVDefUse DU) {
+
// Handle the common case of add<nsw/nuw>
- if (DU.NarrowUse->getOpcode() != Instruction::Add)
- return 0;
+ const unsigned OpCode = DU.NarrowUse->getOpcode();
+ // Only Add/Sub/Mul instructions supported yet.
+ if (OpCode != Instruction::Add && OpCode != Instruction::Sub &&
+ OpCode != Instruction::Mul)
+ return nullptr;
// One operand (NarrowDef) has already been extended to WideDef. Now determine
// if extending the other will lead to a recurrence.
- unsigned ExtendOperIdx = DU.NarrowUse->getOperand(0) == DU.NarrowDef ? 1 : 0;
+ const unsigned ExtendOperIdx =
+ DU.NarrowUse->getOperand(0) == DU.NarrowDef ? 1 : 0;
assert(DU.NarrowUse->getOperand(1-ExtendOperIdx) == DU.NarrowDef && "bad DU");
- const SCEV *ExtendOperExpr = 0;
+ const SCEV *ExtendOperExpr = nullptr;
const OverflowingBinaryOperator *OBO =
cast<OverflowingBinaryOperator>(DU.NarrowUse);
if (IsSigned && OBO->hasNoSignedWrap())
ExtendOperExpr = SE->getZeroExtendExpr(
SE->getSCEV(DU.NarrowUse->getOperand(ExtendOperIdx)), WideType);
else
- return 0;
+ return nullptr;
- // When creating this AddExpr, don't apply the current operations NSW or NUW
+ // When creating this SCEV expr, don't apply the current operations NSW or NUW
// flags. This instruction may be guarded by control flow that the no-wrap
// behavior depends on. Non-control-equivalent instructions can be mapped to
// the same SCEV expression, and it would be incorrect to transfer NSW/NUW
// semantics to those operations.
- const SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(
- SE->getAddExpr(SE->getSCEV(DU.WideDef), ExtendOperExpr));
+ const SCEV *lhs = SE->getSCEV(DU.WideDef);
+ const SCEV *rhs = ExtendOperExpr;
+
+ // Let's swap operands to the initial order for the case of non-commutative
+ // operations, like SUB. See PR21014.
+ if (ExtendOperIdx == 0)
+ std::swap(lhs, rhs);
+ const SCEVAddRecExpr *AddRec =
+ dyn_cast<SCEVAddRecExpr>(getSCEVByOpCode(lhs, rhs, OpCode));
if (!AddRec || AddRec->getLoop() != L)
- return 0;
+ return nullptr;
return AddRec;
}
-/// GetWideRecurrence - Is this instruction potentially interesting from
-/// IVUsers' perspective after widening it's type? In other words, can the
-/// extend be safely hoisted out of the loop with SCEV reducing the value to a
-/// recurrence on the same loop. If so, return the sign or zero extended
-/// recurrence. Otherwise return NULL.
-const SCEVAddRecExpr *WidenIV::GetWideRecurrence(Instruction *NarrowUse) {
+/// Is this instruction potentially interesting for further simplification after
+/// widening it's type? In other words, can the extend be safely hoisted out of
+/// the loop with SCEV reducing the value to a recurrence on the same loop. If
+/// so, return the sign or zero extended recurrence. Otherwise return NULL.
+const SCEVAddRecExpr *WidenIV::getWideRecurrence(Instruction *NarrowUse) {
if (!SE->isSCEVable(NarrowUse->getType()))
- return 0;
+ return nullptr;
const SCEV *NarrowExpr = SE->getSCEV(NarrowUse);
if (SE->getTypeSizeInBits(NarrowExpr->getType())
>= SE->getTypeSizeInBits(WideType)) {
// NarrowUse implicitly widens its operand. e.g. a gep with a narrow
// index. So don't follow this use.
- return 0;
+ return nullptr;
}
const SCEV *WideExpr = IsSigned ?
SE->getZeroExtendExpr(NarrowExpr, WideType);
const SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(WideExpr);
if (!AddRec || AddRec->getLoop() != L)
- return 0;
+ return nullptr;
return AddRec;
}
-/// WidenIVUse - Determine whether an individual user of the narrow IV can be
-/// widened. If so, return the wide clone of the user.
-Instruction *WidenIV::WidenIVUse(NarrowIVDefUse DU, SCEVExpander &Rewriter) {
+/// This IV user cannot be widen. Replace this use of the original narrow IV
+/// with a truncation of the new wide IV to isolate and eliminate the narrow IV.
+static void truncateIVUse(NarrowIVDefUse DU, DominatorTree *DT, LoopInfo *LI) {
+ DEBUG(dbgs() << "INDVARS: Truncate IV " << *DU.WideDef
+ << " for user " << *DU.NarrowUse << "\n");
+ IRBuilder<> Builder(
+ getInsertPointForUses(DU.NarrowUse, DU.NarrowDef, DT, LI));
+ Value *Trunc = Builder.CreateTrunc(DU.WideDef, DU.NarrowDef->getType());
+ DU.NarrowUse->replaceUsesOfWith(DU.NarrowDef, Trunc);
+}
- // Stop traversing the def-use chain at inner-loop phis or post-loop phis.
- if (isa<PHINode>(DU.NarrowUse) &&
- LI->getLoopFor(DU.NarrowUse->getParent()) != L)
- return 0;
+/// If the narrow use is a compare instruction, then widen the compare
+// (and possibly the other operand). The extend operation is hoisted into the
+// loop preheader as far as possible.
+bool WidenIV::widenLoopCompare(NarrowIVDefUse DU) {
+ ICmpInst *Cmp = dyn_cast<ICmpInst>(DU.NarrowUse);
+ if (!Cmp)
+ return false;
+
+ // We can legally widen the comparison in the following two cases:
+ //
+ // - The signedness of the IV extension and comparison match
+ //
+ // - The narrow IV is always positive (and thus its sign extension is equal
+ // to its zero extension). For instance, let's say we're zero extending
+ // %narrow for the following use
+ //
+ // icmp slt i32 %narrow, %val ... (A)
+ //
+ // and %narrow is always positive. Then
+ //
+ // (A) == icmp slt i32 sext(%narrow), sext(%val)
+ // == icmp slt i32 zext(%narrow), sext(%val)
+
+ if (!(DU.NeverNegative || IsSigned == Cmp->isSigned()))
+ return false;
+
+ Value *Op = Cmp->getOperand(Cmp->getOperand(0) == DU.NarrowDef ? 1 : 0);
+ unsigned CastWidth = SE->getTypeSizeInBits(Op->getType());
+ unsigned IVWidth = SE->getTypeSizeInBits(WideType);
+ assert (CastWidth <= IVWidth && "Unexpected width while widening compare.");
+
+ // Widen the compare instruction.
+ IRBuilder<> Builder(
+ getInsertPointForUses(DU.NarrowUse, DU.NarrowDef, DT, LI));
+ DU.NarrowUse->replaceUsesOfWith(DU.NarrowDef, DU.WideDef);
+
+ // Widen the other operand of the compare, if necessary.
+ if (CastWidth < IVWidth) {
+ Value *ExtOp = createExtendInst(Op, WideType, Cmp->isSigned(), Cmp);
+ DU.NarrowUse->replaceUsesOfWith(Op, ExtOp);
+ }
+ return true;
+}
+/// Determine whether an individual user of the narrow IV can be widened. If so,
+/// return the wide clone of the user.
+Instruction *WidenIV::widenIVUse(NarrowIVDefUse DU, SCEVExpander &Rewriter) {
+
+ // Stop traversing the def-use chain at inner-loop phis or post-loop phis.
+ if (PHINode *UsePhi = dyn_cast<PHINode>(DU.NarrowUse)) {
+ if (LI->getLoopFor(UsePhi->getParent()) != L) {
+ // For LCSSA phis, sink the truncate outside the loop.
+ // After SimplifyCFG most loop exit targets have a single predecessor.
+ // Otherwise fall back to a truncate within the loop.
+ if (UsePhi->getNumOperands() != 1)
+ truncateIVUse(DU, DT, LI);
+ else {
+ PHINode *WidePhi =
+ PHINode::Create(DU.WideDef->getType(), 1, UsePhi->getName() + ".wide",
+ UsePhi);
+ WidePhi->addIncoming(DU.WideDef, UsePhi->getIncomingBlock(0));
+ IRBuilder<> Builder(&*WidePhi->getParent()->getFirstInsertionPt());
+ Value *Trunc = Builder.CreateTrunc(WidePhi, DU.NarrowDef->getType());
+ UsePhi->replaceAllUsesWith(Trunc);
+ DeadInsts.emplace_back(UsePhi);
+ DEBUG(dbgs() << "INDVARS: Widen lcssa phi " << *UsePhi
+ << " to " << *WidePhi << "\n");
+ }
+ return nullptr;
+ }
+ }
// Our raison d'etre! Eliminate sign and zero extension.
if (IsSigned ? isa<SExtInst>(DU.NarrowUse) : isa<ZExtInst>(DU.NarrowUse)) {
Value *NewDef = DU.WideDef;
<< " replaced by " << *DU.WideDef << "\n");
++NumElimExt;
DU.NarrowUse->replaceAllUsesWith(NewDef);
- DeadInsts.push_back(DU.NarrowUse);
+ DeadInsts.emplace_back(DU.NarrowUse);
}
// Now that the extend is gone, we want to expose it's uses for potential
// further simplification. We don't need to directly inform SimplifyIVUsers
// push the uses of WideDef here.
// No further widening is needed. The deceased [sz]ext had done it for us.
- return 0;
+ return nullptr;
}
// Does this user itself evaluate to a recurrence after widening?
- const SCEVAddRecExpr *WideAddRec = GetWideRecurrence(DU.NarrowUse);
- if (!WideAddRec) {
- WideAddRec = GetExtendedOperandRecurrence(DU);
- }
+ const SCEVAddRecExpr *WideAddRec = getWideRecurrence(DU.NarrowUse);
+ if (!WideAddRec)
+ WideAddRec = getExtendedOperandRecurrence(DU);
+
if (!WideAddRec) {
+ // If use is a loop condition, try to promote the condition instead of
+ // truncating the IV first.
+ if (widenLoopCompare(DU))
+ return nullptr;
+
// This user does not evaluate to a recurence after widening, so don't
// follow it. Instead insert a Trunc to kill off the original use,
// eventually isolating the original narrow IV so it can be removed.
- IRBuilder<> Builder(getInsertPointForUses(DU.NarrowUse, DU.NarrowDef, DT));
- Value *Trunc = Builder.CreateTrunc(DU.WideDef, DU.NarrowDef->getType());
- DU.NarrowUse->replaceUsesOfWith(DU.NarrowDef, Trunc);
- return 0;
+ truncateIVUse(DU, DT, LI);
+ return nullptr;
}
// Assume block terminators cannot evaluate to a recurrence. We can't to
// insert a Trunc after a terminator if there happens to be a critical edge.
// Reuse the IV increment that SCEVExpander created as long as it dominates
// NarrowUse.
- Instruction *WideUse = 0;
+ Instruction *WideUse = nullptr;
if (WideAddRec == WideIncExpr
&& Rewriter.hoistIVInc(WideInc, DU.NarrowUse))
WideUse = WideInc;
else {
- WideUse = CloneIVUser(DU);
+ WideUse = cloneIVUser(DU, WideAddRec);
if (!WideUse)
- return 0;
+ return nullptr;
}
// Evaluation of WideAddRec ensured that the narrow expression could be
// extended outside the loop without overflow. This suggests that the wide use
if (WideAddRec != SE->getSCEV(WideUse)) {
DEBUG(dbgs() << "Wide use expression mismatch: " << *WideUse
<< ": " << *SE->getSCEV(WideUse) << " != " << *WideAddRec << "\n");
- DeadInsts.push_back(WideUse);
- return 0;
+ DeadInsts.emplace_back(WideUse);
+ return nullptr;
}
// Returning WideUse pushes it on the worklist.
return WideUse;
}
-/// pushNarrowIVUsers - Add eligible users of NarrowDef to NarrowIVUsers.
+/// Add eligible users of NarrowDef to NarrowIVUsers.
///
void WidenIV::pushNarrowIVUsers(Instruction *NarrowDef, Instruction *WideDef) {
- for (Value::use_iterator UI = NarrowDef->use_begin(),
- UE = NarrowDef->use_end(); UI != UE; ++UI) {
- Instruction *NarrowUse = cast<Instruction>(*UI);
+ const SCEV *NarrowSCEV = SE->getSCEV(NarrowDef);
+ bool NeverNegative =
+ SE->isKnownPredicate(ICmpInst::ICMP_SGE, NarrowSCEV,
+ SE->getConstant(NarrowSCEV->getType(), 0));
+ for (User *U : NarrowDef->users()) {
+ Instruction *NarrowUser = cast<Instruction>(U);
// Handle data flow merges and bizarre phi cycles.
- if (!Widened.insert(NarrowUse))
+ if (!Widened.insert(NarrowUser).second)
continue;
- NarrowIVUsers.push_back(NarrowIVDefUse(NarrowDef, NarrowUse, WideDef));
+ NarrowIVUsers.push_back(
+ NarrowIVDefUse(NarrowDef, NarrowUser, WideDef, NeverNegative));
}
}
-/// CreateWideIV - Process a single induction variable. First use the
-/// SCEVExpander to create a wide induction variable that evaluates to the same
-/// recurrence as the original narrow IV. Then use a worklist to forward
-/// traverse the narrow IV's def-use chain. After WidenIVUse has processed all
-/// interesting IV users, the narrow IV will be isolated for removal by
-/// DeleteDeadPHIs.
+/// Process a single induction variable. First use the SCEVExpander to create a
+/// wide induction variable that evaluates to the same recurrence as the
+/// original narrow IV. Then use a worklist to forward traverse the narrow IV's
+/// def-use chain. After widenIVUse has processed all interesting IV users, the
+/// narrow IV will be isolated for removal by DeleteDeadPHIs.
///
/// It would be simpler to delete uses as they are processed, but we must avoid
/// invalidating SCEV expressions.
///
-PHINode *WidenIV::CreateWideIV(SCEVExpander &Rewriter) {
+PHINode *WidenIV::createWideIV(SCEVExpander &Rewriter) {
// Is this phi an induction variable?
const SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(SE->getSCEV(OrigPhi));
if (!AddRec)
- return NULL;
+ return nullptr;
// Widen the induction variable expression.
const SCEV *WideIVExpr = IsSigned ?
// Can the IV be extended outside the loop without overflow?
AddRec = dyn_cast<SCEVAddRecExpr>(WideIVExpr);
if (!AddRec || AddRec->getLoop() != L)
- return NULL;
+ return nullptr;
// An AddRec must have loop-invariant operands. Since this AddRec is
// materialized by a loop header phi, the expression cannot have any post-loop
// either find an existing phi or materialize a new one. Either way, we
// expect a well-formed cyclic phi-with-increments. i.e. any operand not part
// of the phi-SCC dominates the loop entry.
- Instruction *InsertPt = L->getHeader()->begin();
+ Instruction *InsertPt = &L->getHeader()->front();
WidePhi = cast<PHINode>(Rewriter.expandCodeFor(AddRec, WideType, InsertPt));
// Remembering the WideIV increment generated by SCEVExpander allows
- // WidenIVUse to reuse it when widening the narrow IV's increment. We don't
+ // widenIVUse to reuse it when widening the narrow IV's increment. We don't
// employ a general reuse mechanism because the call above is the only call to
// SCEVExpander. Henceforth, we produce 1-to-1 narrow to wide uses.
if (BasicBlock *LatchBlock = L->getLoopLatch()) {
// Process a def-use edge. This may replace the use, so don't hold a
// use_iterator across it.
- Instruction *WideUse = WidenIVUse(DU, Rewriter);
+ Instruction *WideUse = widenIVUse(DU, Rewriter);
// Follow all def-use edges from the previous narrow use.
if (WideUse)
pushNarrowIVUsers(DU.NarrowUse, WideUse);
- // WidenIVUse may have removed the def-use edge.
+ // widenIVUse may have removed the def-use edge.
if (DU.NarrowDef->use_empty())
- DeadInsts.push_back(DU.NarrowDef);
+ DeadInsts.emplace_back(DU.NarrowDef);
}
return WidePhi;
}
+//===----------------------------------------------------------------------===//
+// Live IV Reduction - Minimize IVs live across the loop.
+//===----------------------------------------------------------------------===//
+
+
//===----------------------------------------------------------------------===//
// Simplification of IV users based on SCEV evaluation.
//===----------------------------------------------------------------------===//
+namespace {
+class IndVarSimplifyVisitor : public IVVisitor {
+ ScalarEvolution *SE;
+ const TargetTransformInfo *TTI;
+ PHINode *IVPhi;
+
+public:
+ WideIVInfo WI;
+
+ IndVarSimplifyVisitor(PHINode *IV, ScalarEvolution *SCEV,
+ const TargetTransformInfo *TTI,
+ const DominatorTree *DTree)
+ : SE(SCEV), TTI(TTI), IVPhi(IV) {
+ DT = DTree;
+ WI.NarrowIV = IVPhi;
+ if (ReduceLiveIVs)
+ setSplitOverflowIntrinsics();
+ }
-/// SimplifyAndExtend - Iteratively perform simplification on a worklist of IV
-/// users. Each successive simplification may push more users which may
-/// themselves be candidates for simplification.
+ // Implement the interface used by simplifyUsersOfIV.
+ void visitCast(CastInst *Cast) override { visitIVCast(Cast, WI, SE, TTI); }
+};
+}
+
+/// Iteratively perform simplification on a worklist of IV users. Each
+/// successive simplification may push more users which may themselves be
+/// candidates for simplification.
///
/// Sign/Zero extend elimination is interleaved with IV simplification.
///
-void IndVarSimplify::SimplifyAndExtend(Loop *L,
+void IndVarSimplify::simplifyAndExtend(Loop *L,
SCEVExpander &Rewriter,
- LPPassManager &LPM) {
+ LoopInfo *LI) {
SmallVector<WideIVInfo, 8> WideIVs;
SmallVector<PHINode*, 8> LoopPhis;
// extension. The first time SCEV attempts to normalize sign/zero extension,
// the result becomes final. So for the most predictable results, we delay
// evaluation of sign/zero extend evaluation until needed, and avoid running
- // other SCEV based analysis prior to SimplifyAndExtend.
+ // other SCEV based analysis prior to simplifyAndExtend.
do {
PHINode *CurrIV = LoopPhis.pop_back_val();
// Information about sign/zero extensions of CurrIV.
- WideIVVisitor WIV(CurrIV, SE, TD);
+ IndVarSimplifyVisitor Visitor(CurrIV, SE, TTI, DT);
- Changed |= simplifyUsersOfIV(CurrIV, SE, &LPM, DeadInsts, &WIV);
+ Changed |= simplifyUsersOfIV(CurrIV, SE, DT, LI, DeadInsts, &Visitor);
- if (WIV.WI.WidestNativeType) {
- WideIVs.push_back(WIV.WI);
+ if (Visitor.WI.WidestNativeType) {
+ WideIVs.push_back(Visitor.WI);
}
} while(!LoopPhis.empty());
for (; !WideIVs.empty(); WideIVs.pop_back()) {
WidenIV Widener(WideIVs.back(), LI, SE, DT, DeadInsts);
- if (PHINode *WidePhi = Widener.CreateWideIV(Rewriter)) {
+ if (PHINode *WidePhi = Widener.createWideIV(Rewriter)) {
Changed = true;
LoopPhis.push_back(WidePhi);
}
}
//===----------------------------------------------------------------------===//
-// LinearFunctionTestReplace and its kin. Rewrite the loop exit condition.
+// linearFunctionTestReplace and its kin. Rewrite the loop exit condition.
//===----------------------------------------------------------------------===//
-/// Check for expressions that ScalarEvolution generates to compute
-/// BackedgeTakenInfo. If these expressions have not been reduced, then
-/// expanding them may incur additional cost (albeit in the loop preheader).
-static bool isHighCostExpansion(const SCEV *S, BranchInst *BI,
- SmallPtrSet<const SCEV*, 8> &Processed,
- ScalarEvolution *SE) {
- if (!Processed.insert(S))
- return false;
-
- // If the backedge-taken count is a UDiv, it's very likely a UDiv that
- // ScalarEvolution's HowFarToZero or HowManyLessThans produced to compute a
- // precise expression, rather than a UDiv from the user's code. If we can't
- // find a UDiv in the code with some simple searching, assume the former and
- // forego rewriting the loop.
- if (isa<SCEVUDivExpr>(S)) {
- ICmpInst *OrigCond = dyn_cast<ICmpInst>(BI->getCondition());
- if (!OrigCond) return true;
- const SCEV *R = SE->getSCEV(OrigCond->getOperand(1));
- R = SE->getMinusSCEV(R, SE->getConstant(R->getType(), 1));
- if (R != S) {
- const SCEV *L = SE->getSCEV(OrigCond->getOperand(0));
- L = SE->getMinusSCEV(L, SE->getConstant(L->getType(), 1));
- if (L != S)
- return true;
- }
- }
-
- // Recurse past add expressions, which commonly occur in the
- // BackedgeTakenCount. They may already exist in program code, and if not,
- // they are not too expensive rematerialize.
- if (const SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(S)) {
- for (SCEVAddExpr::op_iterator I = Add->op_begin(), E = Add->op_end();
- I != E; ++I) {
- if (isHighCostExpansion(*I, BI, Processed, SE))
- return true;
- }
- return false;
- }
-
- // HowManyLessThans uses a Max expression whenever the loop is not guarded by
- // the exit condition.
- if (isa<SCEVSMaxExpr>(S) || isa<SCEVUMaxExpr>(S))
- return true;
-
- // If we haven't recognized an expensive SCEV pattern, assume it's an
- // expression produced by program code.
- return false;
-}
-
-/// canExpandBackedgeTakenCount - Return true if this loop's backedge taken
-/// count expression can be safely and cheaply expanded into an instruction
-/// sequence that can be used by LinearFunctionTestReplace.
+/// Return true if this loop's backedge taken count expression can be safely and
+/// cheaply expanded into an instruction sequence that can be used by
+/// linearFunctionTestReplace.
///
/// TODO: This fails for pointer-type loop counters with greater than one byte
/// strides, consequently preventing LFTR from running. For the purpose of LFTR
/// used by ABI constrained operation, as opposed to inttoptr/ptrtoint).
/// However, we don't yet have a strong motivation for converting loop tests
/// into inequality tests.
-static bool canExpandBackedgeTakenCount(Loop *L, ScalarEvolution *SE) {
+static bool canExpandBackedgeTakenCount(Loop *L, ScalarEvolution *SE,
+ SCEVExpander &Rewriter) {
const SCEV *BackedgeTakenCount = SE->getBackedgeTakenCount(L);
if (isa<SCEVCouldNotCompute>(BackedgeTakenCount) ||
BackedgeTakenCount->isZero())
return false;
// Can't rewrite non-branch yet.
- BranchInst *BI = dyn_cast<BranchInst>(L->getExitingBlock()->getTerminator());
- if (!BI)
+ if (!isa<BranchInst>(L->getExitingBlock()->getTerminator()))
return false;
- SmallPtrSet<const SCEV*, 8> Processed;
- if (isHighCostExpansion(BackedgeTakenCount, BI, Processed, SE))
+ if (Rewriter.isHighCostExpansion(BackedgeTakenCount, L))
return false;
return true;
}
-/// getLoopPhiForCounter - Return the loop header phi IFF IncV adds a loop
-/// invariant value to the phi.
+/// Return the loop header phi IFF IncV adds a loop invariant value to the phi.
static PHINode *getLoopPhiForCounter(Value *IncV, Loop *L, DominatorTree *DT) {
Instruction *IncI = dyn_cast<Instruction>(IncV);
if (!IncI)
- return 0;
+ return nullptr;
switch (IncI->getOpcode()) {
case Instruction::Add:
if (IncI->getNumOperands() == 2)
break;
default:
- return 0;
+ return nullptr;
}
PHINode *Phi = dyn_cast<PHINode>(IncI->getOperand(0));
if (Phi && Phi->getParent() == L->getHeader()) {
if (isLoopInvariant(IncI->getOperand(1), L, DT))
return Phi;
- return 0;
+ return nullptr;
}
if (IncI->getOpcode() == Instruction::GetElementPtr)
- return 0;
+ return nullptr;
// Allow add/sub to be commuted.
Phi = dyn_cast<PHINode>(IncI->getOperand(1));
if (isLoopInvariant(IncI->getOperand(0), L, DT))
return Phi;
}
- return 0;
+ return nullptr;
}
/// Return the compare guarding the loop latch, or NULL for unrecognized tests.
BasicBlock *LatchBlock = L->getLoopLatch();
// Don't bother with LFTR if the loop is not properly simplified.
if (!LatchBlock)
- return 0;
+ return nullptr;
BranchInst *BI = dyn_cast<BranchInst>(L->getExitingBlock()->getTerminator());
assert(BI && "expected exit branch");
return dyn_cast<ICmpInst>(BI->getCondition());
}
-/// needsLFTR - LinearFunctionTestReplace policy. Return true unless we can show
-/// that the current exit test is already sufficiently canonical.
+/// linearFunctionTestReplace policy. Return true unless we can show that the
+/// current exit test is already sufficiently canonical.
static bool needsLFTR(Loop *L, DominatorTree *DT) {
// Do LFTR to simplify the exit condition to an ICMP.
ICmpInst *Cond = getLoopTest(L);
/// Recursive helper for hasConcreteDef(). Unfortunately, this currently boils
/// down to checking that all operands are constant and listing instructions
/// that may hide undef.
-static bool hasConcreteDefImpl(Value *V, SmallPtrSet<Value*, 8> &Visited,
+static bool hasConcreteDefImpl(Value *V, SmallPtrSetImpl<Value*> &Visited,
unsigned Depth) {
if (isa<Constant>(V))
return !isa<UndefValue>(V);
return false;
// Optimistically handle other instructions.
- for (User::op_iterator OI = I->op_begin(), E = I->op_end(); OI != E; ++OI) {
- if (!Visited.insert(*OI))
+ for (Value *Op : I->operands()) {
+ if (!Visited.insert(Op).second)
continue;
- if (!hasConcreteDefImpl(*OI, Visited, Depth+1))
+ if (!hasConcreteDefImpl(Op, Visited, Depth+1))
return false;
}
return true;
return hasConcreteDefImpl(V, Visited, 0);
}
-/// AlmostDeadIV - Return true if this IV has any uses other than the (soon to
-/// be rewritten) loop exit test.
+/// Return true if this IV has any uses other than the (soon to be rewritten)
+/// loop exit test.
static bool AlmostDeadIV(PHINode *Phi, BasicBlock *LatchBlock, Value *Cond) {
int LatchIdx = Phi->getBasicBlockIndex(LatchBlock);
Value *IncV = Phi->getIncomingValue(LatchIdx);
- for (Value::use_iterator UI = Phi->use_begin(), UE = Phi->use_end();
- UI != UE; ++UI) {
- if (*UI != Cond && *UI != IncV) return false;
- }
+ for (User *U : Phi->users())
+ if (U != Cond && U != IncV) return false;
- for (Value::use_iterator UI = IncV->use_begin(), UE = IncV->use_end();
- UI != UE; ++UI) {
- if (*UI != Cond && *UI != Phi) return false;
- }
+ for (User *U : IncV->users())
+ if (U != Cond && U != Phi) return false;
return true;
}
-/// FindLoopCounter - Find an affine IV in canonical form.
+/// Find an affine IV in canonical form.
///
/// BECount may be an i8* pointer type. The pointer difference is already
/// valid count without scaling the address stride, so it remains a pointer
/// FIXME: Accept non-unit stride as long as SCEV can reduce BECount * Stride.
/// This is difficult in general for SCEV because of potential overflow. But we
/// could at least handle constant BECounts.
-static PHINode *
-FindLoopCounter(Loop *L, const SCEV *BECount,
- ScalarEvolution *SE, DominatorTree *DT, const DataLayout *TD) {
+static PHINode *FindLoopCounter(Loop *L, const SCEV *BECount,
+ ScalarEvolution *SE, DominatorTree *DT) {
uint64_t BCWidth = SE->getTypeSizeInBits(BECount->getType());
Value *Cond =
cast<BranchInst>(L->getExitingBlock()->getTerminator())->getCondition();
// Loop over all of the PHI nodes, looking for a simple counter.
- PHINode *BestPhi = 0;
- const SCEV *BestInit = 0;
+ PHINode *BestPhi = nullptr;
+ const SCEV *BestInit = nullptr;
BasicBlock *LatchBlock = L->getLoopLatch();
assert(LatchBlock && "needsLFTR should guarantee a loop latch");
// AR may be wider than BECount. With eq/ne tests overflow is immaterial.
// AR may not be a narrower type, or we may never exit.
uint64_t PhiWidth = SE->getTypeSizeInBits(AR->getType());
- if (PhiWidth < BCWidth || (TD && !TD->isLegalInteger(PhiWidth)))
+ if (PhiWidth < BCWidth ||
+ !L->getHeader()->getModule()->getDataLayout().isLegalInteger(PhiWidth))
continue;
const SCEV *Step = dyn_cast<SCEVConstant>(AR->getStepRecurrence(*SE));
return BestPhi;
}
-/// genLoopLimit - Help LinearFunctionTestReplace by generating a value that
-/// holds the RHS of the new loop test.
+/// Help linearFunctionTestReplace by generating a value that holds the RHS of
+/// the new loop test.
static Value *genLoopLimit(PHINode *IndVar, const SCEV *IVCount, Loop *L,
SCEVExpander &Rewriter, ScalarEvolution *SE) {
const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(SE->getSCEV(IndVar));
&& "unit stride pointer IV must be i8*");
IRBuilder<> Builder(L->getLoopPreheader()->getTerminator());
- return Builder.CreateGEP(GEPBase, GEPOffset, "lftr.limit");
+ return Builder.CreateGEP(nullptr, GEPBase, GEPOffset, "lftr.limit");
}
else {
// In any other case, convert both IVInit and IVCount to integers before
// IVInit integer and IVCount pointer would only occur if a canonical IV
// were generated on top of case #2, which is not expected.
- const SCEV *IVLimit = 0;
+ const SCEV *IVLimit = nullptr;
// For unit stride, IVCount = Start + BECount with 2's complement overflow.
// For non-zero Start, compute IVCount here.
if (AR->getStart()->isZero())
}
}
-/// LinearFunctionTestReplace - This method rewrites the exit condition of the
-/// loop to be a canonical != comparison against the incremented loop induction
-/// variable. This pass is able to rewrite the exit tests of any loop where the
-/// SCEV analysis can determine a loop-invariant trip count of the loop, which
-/// is actually a much broader range than just linear tests.
+/// This method rewrites the exit condition of the loop to be a canonical !=
+/// comparison against the incremented loop induction variable. This pass is
+/// able to rewrite the exit tests of any loop where the SCEV analysis can
+/// determine a loop-invariant trip count of the loop, which is actually a much
+/// broader range than just linear tests.
Value *IndVarSimplify::
-LinearFunctionTestReplace(Loop *L,
+linearFunctionTestReplace(Loop *L,
const SCEV *BackedgeTakenCount,
PHINode *IndVar,
SCEVExpander &Rewriter) {
- assert(canExpandBackedgeTakenCount(L, SE) && "precondition");
+ assert(canExpandBackedgeTakenCount(L, SE, Rewriter) && "precondition");
// Initialize CmpIndVar and IVCount to their preincremented values.
Value *CmpIndVar = IndVar;
// This addition may overflow, which is valid as long as the comparison is
// truncated to BackedgeTakenCount->getType().
IVCount = SE->getAddExpr(BackedgeTakenCount,
- SE->getConstant(BackedgeTakenCount->getType(), 1));
+ SE->getOne(BackedgeTakenCount->getType()));
// The BackedgeTaken expression contains the number of times that the
// backedge branches to the loop header. This is one less than the
// number of times the loop executes, so use the incremented indvar.
const SCEV *ARStep = AR->getStepRecurrence(*SE);
// For constant IVCount, avoid truncation.
if (isa<SCEVConstant>(ARStart) && isa<SCEVConstant>(IVCount)) {
- const APInt &Start = cast<SCEVConstant>(ARStart)->getValue()->getValue();
- APInt Count = cast<SCEVConstant>(IVCount)->getValue()->getValue();
+ const APInt &Start = cast<SCEVConstant>(ARStart)->getAPInt();
+ APInt Count = cast<SCEVConstant>(IVCount)->getAPInt();
// Note that the post-inc value of BackedgeTakenCount may have overflowed
// above such that IVCount is now zero.
if (IVCount != BackedgeTakenCount && Count == 0) {
}
//===----------------------------------------------------------------------===//
-// SinkUnusedInvariants. A late subpass to cleanup loop preheaders.
+// sinkUnusedInvariants. A late subpass to cleanup loop preheaders.
//===----------------------------------------------------------------------===//
/// If there's a single exit block, sink any loop-invariant values that
/// were defined in the preheader but not used inside the loop into the
/// exit block to reduce register pressure in the loop.
-void IndVarSimplify::SinkUnusedInvariants(Loop *L) {
+void IndVarSimplify::sinkUnusedInvariants(Loop *L) {
BasicBlock *ExitBlock = L->getExitBlock();
if (!ExitBlock) return;
BasicBlock *Preheader = L->getLoopPreheader();
if (!Preheader) return;
- Instruction *InsertPt = ExitBlock->getFirstInsertionPt();
- BasicBlock::iterator I = Preheader->getTerminator();
+ Instruction *InsertPt = &*ExitBlock->getFirstInsertionPt();
+ BasicBlock::iterator I(Preheader->getTerminator());
while (I != Preheader->begin()) {
--I;
// New instructions were inserted at the end of the preheader.
if (isa<DbgInfoIntrinsic>(I))
continue;
- // Skip landingpad instructions.
- if (isa<LandingPadInst>(I))
+ // Skip eh pad instructions.
+ if (I->isEHPad())
continue;
// Don't sink alloca: we never want to sink static alloca's out of the
// Determine if there is a use in or before the loop (direct or
// otherwise).
bool UsedInLoop = false;
- for (Value::use_iterator UI = I->use_begin(), UE = I->use_end();
- UI != UE; ++UI) {
- User *U = *UI;
- BasicBlock *UseBB = cast<Instruction>(U)->getParent();
- if (PHINode *P = dyn_cast<PHINode>(U)) {
+ for (Use &U : I->uses()) {
+ Instruction *User = cast<Instruction>(U.getUser());
+ BasicBlock *UseBB = User->getParent();
+ if (PHINode *P = dyn_cast<PHINode>(User)) {
unsigned i =
- PHINode::getIncomingValueNumForOperand(UI.getOperandNo());
+ PHINode::getIncomingValueNumForOperand(U.getOperandNo());
UseBB = P->getIncomingBlock(i);
}
if (UseBB == Preheader || L->contains(UseBB)) {
continue;
// Otherwise, sink it to the exit block.
- Instruction *ToMove = I;
+ Instruction *ToMove = &*I;
bool Done = false;
if (I != Preheader->begin()) {
//===----------------------------------------------------------------------===//
bool IndVarSimplify::runOnLoop(Loop *L, LPPassManager &LPM) {
+ if (skipOptnoneFunction(L))
+ return false;
+
// If LoopSimplify form is not available, stay out of trouble. Some notes:
// - LSR currently only supports LoopSimplify-form loops. Indvars'
// canonicalization can be a pessimization without LSR to "clean up"
if (!L->isLoopSimplifyForm())
return false;
- LI = &getAnalysis<LoopInfo>();
- SE = &getAnalysis<ScalarEvolution>();
- DT = &getAnalysis<DominatorTree>();
- TD = getAnalysisIfAvailable<DataLayout>();
- TLI = getAnalysisIfAvailable<TargetLibraryInfo>();
+ LI = &getAnalysis<LoopInfoWrapperPass>().getLoopInfo();
+ SE = &getAnalysis<ScalarEvolutionWrapperPass>().getSE();
+ DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree();
+ auto *TLIP = getAnalysisIfAvailable<TargetLibraryInfoWrapperPass>();
+ TLI = TLIP ? &TLIP->getTLI() : nullptr;
+ auto *TTIP = getAnalysisIfAvailable<TargetTransformInfoWrapperPass>();
+ TTI = TTIP ? &TTIP->getTTI(*L->getHeader()->getParent()) : nullptr;
+ const DataLayout &DL = L->getHeader()->getModule()->getDataLayout();
DeadInsts.clear();
Changed = false;
// If there are any floating-point recurrences, attempt to
// transform them to use integer recurrences.
- RewriteNonIntegerIVs(L);
+ rewriteNonIntegerIVs(L);
const SCEV *BackedgeTakenCount = SE->getBackedgeTakenCount(L);
// Create a rewriter object which we'll use to transform the code with.
- SCEVExpander Rewriter(*SE, "indvars");
+ SCEVExpander Rewriter(*SE, DL, "indvars");
#ifndef NDEBUG
Rewriter.setDebugType(DEBUG_TYPE);
#endif
// other expressions involving loop IVs have been evaluated. This helps SCEV
// set no-wrap flags before normalizing sign/zero extension.
Rewriter.disableCanonicalMode();
- SimplifyAndExtend(L, Rewriter, LPM);
+ simplifyAndExtend(L, Rewriter, LI);
// Check to see if this loop has a computable loop-invariant execution count.
// If so, this means that we can compute the final value of any expressions
// loop into any instructions outside of the loop that use the final values of
// the current expressions.
//
- if (!isa<SCEVCouldNotCompute>(BackedgeTakenCount))
- RewriteLoopExitValues(L, Rewriter);
+ if (ReplaceExitValue != NeverRepl &&
+ !isa<SCEVCouldNotCompute>(BackedgeTakenCount))
+ rewriteLoopExitValues(L, Rewriter);
// Eliminate redundant IV cycles.
NumElimIV += Rewriter.replaceCongruentIVs(L, DT, DeadInsts);
// If we have a trip count expression, rewrite the loop's exit condition
// using it. We can currently only handle loops with a single exit.
- if (canExpandBackedgeTakenCount(L, SE) && needsLFTR(L, DT)) {
- PHINode *IndVar = FindLoopCounter(L, BackedgeTakenCount, SE, DT, TD);
+ if (canExpandBackedgeTakenCount(L, SE, Rewriter) && needsLFTR(L, DT)) {
+ PHINode *IndVar = FindLoopCounter(L, BackedgeTakenCount, SE, DT);
if (IndVar) {
// Check preconditions for proper SCEVExpander operation. SCEV does not
// express SCEVExpander's dependencies, such as LoopSimplify. Instead any
// pass that uses the SCEVExpander must do it. This does not work well for
- // loop passes because SCEVExpander makes assumptions about all loops, while
- // LoopPassManager only forces the current loop to be simplified.
+ // loop passes because SCEVExpander makes assumptions about all loops,
+ // while LoopPassManager only forces the current loop to be simplified.
//
// FIXME: SCEV expansion has no way to bail out, so the caller must
// explicitly check any assumptions made by SCEV. Brittle.
const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(BackedgeTakenCount);
if (!AR || AR->getLoop()->getLoopPreheader())
- (void)LinearFunctionTestReplace(L, BackedgeTakenCount, IndVar,
+ (void)linearFunctionTestReplace(L, BackedgeTakenCount, IndVar,
Rewriter);
}
}
// which are now dead.
while (!DeadInsts.empty())
if (Instruction *Inst =
- dyn_cast_or_null<Instruction>(&*DeadInsts.pop_back_val()))
+ dyn_cast_or_null<Instruction>(DeadInsts.pop_back_val()))
RecursivelyDeleteTriviallyDeadInstructions(Inst, TLI);
// The Rewriter may not be used from this point on.
// Loop-invariant instructions in the preheader that aren't used in the
// loop may be sunk below the loop to reduce register pressure.
- SinkUnusedInvariants(L);
+ sinkUnusedInvariants(L);
// Clean up dead instructions.
Changed |= DeleteDeadPHIs(L->getHeader(), TLI);
+
// Check a post-condition.
- assert(L->isLCSSAForm(*DT) &&
- "Indvars did not leave the loop in lcssa form!");
+ assert(L->isRecursivelyLCSSAForm(*DT) && "Indvars did not preserve LCSSA!");
// Verify that LFTR, and any other change have not interfered with SCEV's
// ability to compute trip count.
projects / oota-llvm.git / blobdiff