+static void deleteDeadInstruction(Instruction *I,
+ const TargetLibraryInfo *TLI) {
+ SmallVector<Value *, 16> Operands(I->value_op_begin(), I->value_op_end());
+ I->replaceAllUsesWith(UndefValue::get(I->getType()));
+ I->eraseFromParent();
+ for (Value *Op : Operands)
+ RecursivelyDeleteTriviallyDeadInstructions(Op, TLI);
+}
+
+//===----------------------------------------------------------------------===//
+//
+// Implementation of NclPopcountRecognize
+//
+//===----------------------------------------------------------------------===//
+
+NclPopcountRecognize::NclPopcountRecognize(LoopIdiomRecognize &TheLIR)
+ : LIR(TheLIR), CurLoop(TheLIR.getLoop()), PreCondBB(nullptr) {}
+
+bool NclPopcountRecognize::preliminaryScreen() {
+ const TargetTransformInfo *TTI = LIR.getTargetTransformInfo();
+ if (TTI->getPopcntSupport(32) != TargetTransformInfo::PSK_FastHardware)
+ return false;
+
+ // Counting population are usually conducted by few arithmetic instructions.
+ // Such instructions can be easilly "absorbed" by vacant slots in a
+ // non-compact loop. Therefore, recognizing popcount idiom only makes sense
+ // in a compact loop.
+
+ // Give up if the loop has multiple blocks or multiple backedges.
+ if (CurLoop->getNumBackEdges() != 1 || CurLoop->getNumBlocks() != 1)
+ return false;
+
+ BasicBlock *LoopBody = *(CurLoop->block_begin());
+ if (LoopBody->size() >= 20) {
+ // The loop is too big, bail out.
+ return false;
+ }
+
+ // It should have a preheader containing nothing but an unconditional branch.
+ BasicBlock *PH = CurLoop->getLoopPreheader();
+ if (!PH)
+ return false;
+ if (&PH->front() != PH->getTerminator())
+ return false;
+ auto *EntryBI = dyn_cast<BranchInst>(PH->getTerminator());
+ if (!EntryBI || EntryBI->isConditional())
+ return false;
+
+ // It should have a precondition block where the generated popcount instrinsic
+ // function can be inserted.
+ PreCondBB = PH->getSinglePredecessor();
+ if (!PreCondBB)
+ return false;
+ auto *PreCondBI = dyn_cast<BranchInst>(PreCondBB->getTerminator());
+ if (!PreCondBI || PreCondBI->isUnconditional())
+ return false;
+
+ return true;
+}
+
+Value *NclPopcountRecognize::matchCondition(BranchInst *Br,
+ BasicBlock *LoopEntry) const {
+ if (!Br || !Br->isConditional())
+ return nullptr;
+
+ ICmpInst *Cond = dyn_cast<ICmpInst>(Br->getCondition());
+ if (!Cond)
+ return nullptr;
+
+ ConstantInt *CmpZero = dyn_cast<ConstantInt>(Cond->getOperand(1));
+ if (!CmpZero || !CmpZero->isZero())
+ return nullptr;
+
+ ICmpInst::Predicate Pred = Cond->getPredicate();
+ if ((Pred == ICmpInst::ICMP_NE && Br->getSuccessor(0) == LoopEntry) ||
+ (Pred == ICmpInst::ICMP_EQ && Br->getSuccessor(1) == LoopEntry))
+ return Cond->getOperand(0);
+
+ return nullptr;
+}
+
+bool NclPopcountRecognize::detectIdiom(Instruction *&CntInst, PHINode *&CntPhi,
+ Value *&Var) const {
+ // Following code tries to detect this idiom:
+ //
+ // if (x0 != 0)
+ // goto loop-exit // the precondition of the loop
+ // cnt0 = init-val;
+ // do {
+ // x1 = phi (x0, x2);
+ // cnt1 = phi(cnt0, cnt2);
+ //
+ // cnt2 = cnt1 + 1;
+ // ...
+ // x2 = x1 & (x1 - 1);
+ // ...
+ // } while(x != 0);
+ //
+ // loop-exit:
+ //
+
+ // step 1: Check to see if the look-back branch match this pattern:
+ // "if (a!=0) goto loop-entry".
+ BasicBlock *LoopEntry;
+ Instruction *DefX2, *CountInst;
+ Value *VarX1, *VarX0;
+ PHINode *PhiX, *CountPhi;
+
+ DefX2 = CountInst = nullptr;
+ VarX1 = VarX0 = nullptr;
+ PhiX = CountPhi = nullptr;
+ LoopEntry = *(CurLoop->block_begin());
+
+ // step 1: Check if the loop-back branch is in desirable form.
+ {
+ if (Value *T = matchCondition(
+ dyn_cast<BranchInst>(LoopEntry->getTerminator()), LoopEntry))
+ DefX2 = dyn_cast<Instruction>(T);
+ else
+ return false;
+ }
+
+ // step 2: detect instructions corresponding to "x2 = x1 & (x1 - 1)"
+ {
+ if (!DefX2 || DefX2->getOpcode() != Instruction::And)
+ return false;
+
+ BinaryOperator *SubOneOp;
+
+ if ((SubOneOp = dyn_cast<BinaryOperator>(DefX2->getOperand(0))))
+ VarX1 = DefX2->getOperand(1);
+ else {
+ VarX1 = DefX2->getOperand(0);
+ SubOneOp = dyn_cast<BinaryOperator>(DefX2->getOperand(1));
+ }
+ if (!SubOneOp)
+ return false;
+
+ Instruction *SubInst = cast<Instruction>(SubOneOp);
+ ConstantInt *Dec = dyn_cast<ConstantInt>(SubInst->getOperand(1));
+ if (!Dec ||
+ !((SubInst->getOpcode() == Instruction::Sub && Dec->isOne()) ||
+ (SubInst->getOpcode() == Instruction::Add &&
+ Dec->isAllOnesValue()))) {
+ return false;
+ }
+ }
+
+ // step 3: Check the recurrence of variable X
+ {
+ PhiX = dyn_cast<PHINode>(VarX1);
+ if (!PhiX ||
+ (PhiX->getOperand(0) != DefX2 && PhiX->getOperand(1) != DefX2)) {
+ return false;
+ }
+ }
+
+ // step 4: Find the instruction which count the population: cnt2 = cnt1 + 1
+ {
+ CountInst = nullptr;
+ for (BasicBlock::iterator Iter = LoopEntry->getFirstNonPHI(),
+ IterE = LoopEntry->end();
+ Iter != IterE; Iter++) {
+ Instruction *Inst = Iter;
+ if (Inst->getOpcode() != Instruction::Add)
+ continue;
+
+ ConstantInt *Inc = dyn_cast<ConstantInt>(Inst->getOperand(1));
+ if (!Inc || !Inc->isOne())
+ continue;
+
+ PHINode *Phi = dyn_cast<PHINode>(Inst->getOperand(0));
+ if (!Phi || Phi->getParent() != LoopEntry)
+ continue;
+
+ // Check if the result of the instruction is live of the loop.
+ bool LiveOutLoop = false;
+ for (User *U : Inst->users()) {
+ if ((cast<Instruction>(U))->getParent() != LoopEntry) {
+ LiveOutLoop = true;
+ break;
+ }
+ }
+
+ if (LiveOutLoop) {
+ CountInst = Inst;
+ CountPhi = Phi;
+ break;
+ }
+ }
+
+ if (!CountInst)
+ return false;
+ }
+
+ // step 5: check if the precondition is in this form:
+ // "if (x != 0) goto loop-head ; else goto somewhere-we-don't-care;"
+ {
+ auto *PreCondBr = dyn_cast<BranchInst>(PreCondBB->getTerminator());
+ Value *T = matchCondition(PreCondBr, CurLoop->getLoopPreheader());
+ if (T != PhiX->getOperand(0) && T != PhiX->getOperand(1))
+ return false;
+
+ CntInst = CountInst;
+ CntPhi = CountPhi;
+ Var = T;
+ }
+
+ return true;
+}
+
+void NclPopcountRecognize::transform(Instruction *CntInst, PHINode *CntPhi,
+ Value *Var) {
+
+ ScalarEvolution *SE = LIR.getScalarEvolution();
+ TargetLibraryInfo *TLI = LIR.getTargetLibraryInfo();
+ BasicBlock *PreHead = CurLoop->getLoopPreheader();
+ auto *PreCondBr = dyn_cast<BranchInst>(PreCondBB->getTerminator());
+ const DebugLoc DL = CntInst->getDebugLoc();
+
+ // Assuming before transformation, the loop is following:
+ // if (x) // the precondition
+ // do { cnt++; x &= x - 1; } while(x);
+
+ // Step 1: Insert the ctpop instruction at the end of the precondition block
+ IRBuilderTy Builder(PreCondBr);
+ Value *PopCnt, *PopCntZext, *NewCount, *TripCnt;
+ {
+ PopCnt = createPopcntIntrinsic(Builder, Var, DL);
+ NewCount = PopCntZext =
+ Builder.CreateZExtOrTrunc(PopCnt, cast<IntegerType>(CntPhi->getType()));
+
+ if (NewCount != PopCnt)
+ (cast<Instruction>(NewCount))->setDebugLoc(DL);
+
+ // TripCnt is exactly the number of iterations the loop has
+ TripCnt = NewCount;
+
+ // If the population counter's initial value is not zero, insert Add Inst.
+ Value *CntInitVal = CntPhi->getIncomingValueForBlock(PreHead);
+ ConstantInt *InitConst = dyn_cast<ConstantInt>(CntInitVal);
+ if (!InitConst || !InitConst->isZero()) {
+ NewCount = Builder.CreateAdd(NewCount, CntInitVal);
+ (cast<Instruction>(NewCount))->setDebugLoc(DL);