C++ IntrinsicInst::replaceAllUsesWith方法代码示例

bool AMDGPUCodeGenPrepare::promoteUniformBitreverseToI32(
    IntrinsicInst &I) const {
  assert(I.getIntrinsicID() == Intrinsic::bitreverse &&
         "I must be bitreverse intrinsic");
  assert(needsPromotionToI32(I.getType()) &&
         "I does not need promotion to i32");

  IRBuilder<> Builder(&I);
  Builder.SetCurrentDebugLocation(I.getDebugLoc());

  Type *I32Ty = getI32Ty(Builder, I.getType());
  Function *I32 =
      Intrinsic::getDeclaration(Mod, Intrinsic::bitreverse, { I32Ty });
  Value *ExtOp = Builder.CreateZExt(I.getOperand(0), I32Ty);
  Value *ExtRes = Builder.CreateCall(I32, { ExtOp });
  Value *LShrOp =
      Builder.CreateLShr(ExtRes, 32 - getBaseElementBitWidth(I.getType()));
  Value *TruncRes =
      Builder.CreateTrunc(LShrOp, I.getType());

  I.replaceAllUsesWith(TruncRes);
  I.eraseFromParent();

  return true;
}

void AMDGPUPromoteAlloca::handleAlloca(AllocaInst &I) {
  // Array allocations are probably not worth handling, since an allocation of
  // the array type is the canonical form.
  if (!I.isStaticAlloca() || I.isArrayAllocation())
    return;

  IRBuilder<> Builder(&I);

  // First try to replace the alloca with a vector
  Type *AllocaTy = I.getAllocatedType();

  DEBUG(dbgs() << "Trying to promote " << I << '\n');

  if (tryPromoteAllocaToVector(&I))
    return;

  DEBUG(dbgs() << " alloca is not a candidate for vectorization.\n");

  const Function &ContainingFunction = *I.getParent()->getParent();

  // FIXME: We should also try to get this value from the reqd_work_group_size
  // function attribute if it is available.
  unsigned WorkGroupSize = AMDGPU::getMaximumWorkGroupSize(ContainingFunction);

  int AllocaSize =
      WorkGroupSize * Mod->getDataLayout().getTypeAllocSize(AllocaTy);

  if (AllocaSize > LocalMemAvailable) {
    DEBUG(dbgs() << " Not enough local memory to promote alloca.\n");
    return;
  }

  std::vector<Value*> WorkList;

  if (!collectUsesWithPtrTypes(&I, WorkList)) {
    DEBUG(dbgs() << " Do not know how to convert all uses\n");
    return;
  }

  DEBUG(dbgs() << "Promoting alloca to local memory\n");
  LocalMemAvailable -= AllocaSize;

  Function *F = I.getParent()->getParent();

  Type *GVTy = ArrayType::get(I.getAllocatedType(), WorkGroupSize);
  GlobalVariable *GV = new GlobalVariable(
      *Mod, GVTy, false, GlobalValue::InternalLinkage,
      UndefValue::get(GVTy),
      Twine(F->getName()) + Twine('.') + I.getName(),
      nullptr,
      GlobalVariable::NotThreadLocal,
      AMDGPUAS::LOCAL_ADDRESS);
  GV->setUnnamedAddr(true);
  GV->setAlignment(I.getAlignment());

  Value *TCntY, *TCntZ;

  std::tie(TCntY, TCntZ) = getLocalSizeYZ(Builder);
  Value *TIdX = getWorkitemID(Builder, 0);
  Value *TIdY = getWorkitemID(Builder, 1);
  Value *TIdZ = getWorkitemID(Builder, 2);

  Value *Tmp0 = Builder.CreateMul(TCntY, TCntZ, "", true, true);
  Tmp0 = Builder.CreateMul(Tmp0, TIdX);
  Value *Tmp1 = Builder.CreateMul(TIdY, TCntZ, "", true, true);
  Value *TID = Builder.CreateAdd(Tmp0, Tmp1);
  TID = Builder.CreateAdd(TID, TIdZ);

  Value *Indices[] = {
    Constant::getNullValue(Type::getInt32Ty(Mod->getContext())),
    TID
  };

  Value *Offset = Builder.CreateInBoundsGEP(GVTy, GV, Indices);
  I.mutateType(Offset->getType());
  I.replaceAllUsesWith(Offset);
  I.eraseFromParent();

  for (Value *V : WorkList) {
    CallInst *Call = dyn_cast<CallInst>(V);
    if (!Call) {
      Type *EltTy = V->getType()->getPointerElementType();
      PointerType *NewTy = PointerType::get(EltTy, AMDGPUAS::LOCAL_ADDRESS);

      // The operand's value should be corrected on its own.
      if (isa<AddrSpaceCastInst>(V))
        continue;

      // FIXME: It doesn't really make sense to try to do this for all
      // instructions.
      V->mutateType(NewTy);
      continue;
    }

    IntrinsicInst *Intr = dyn_cast<IntrinsicInst>(Call);
    if (!Intr) {
      // FIXME: What is this for? It doesn't make sense to promote arbitrary
      // function calls. If the call is to a defined function that can also be
      // promoted, we should be able to do this once that function is also
      // rewritten.
//.........这里部分代码省略.........

//.........这里部分代码省略.........
  Function *F = I.getParent()->getParent();

  Type *GVTy = ArrayType::get(I.getAllocatedType(), WorkGroupSize);
  GlobalVariable *GV = new GlobalVariable(
      *Mod, GVTy, false, GlobalValue::InternalLinkage,
      UndefValue::get(GVTy),
      Twine(F->getName()) + Twine('.') + I.getName(),
      nullptr,
      GlobalVariable::NotThreadLocal,
      AS.LOCAL_ADDRESS);
  GV->setUnnamedAddr(GlobalValue::UnnamedAddr::Global);
  GV->setAlignment(I.getAlignment());

  Value *TCntY, *TCntZ;

  std::tie(TCntY, TCntZ) = getLocalSizeYZ(Builder);
  Value *TIdX = getWorkitemID(Builder, 0);
  Value *TIdY = getWorkitemID(Builder, 1);
  Value *TIdZ = getWorkitemID(Builder, 2);

  Value *Tmp0 = Builder.CreateMul(TCntY, TCntZ, "", true, true);
  Tmp0 = Builder.CreateMul(Tmp0, TIdX);
  Value *Tmp1 = Builder.CreateMul(TIdY, TCntZ, "", true, true);
  Value *TID = Builder.CreateAdd(Tmp0, Tmp1);
  TID = Builder.CreateAdd(TID, TIdZ);

  Value *Indices[] = {
    Constant::getNullValue(Type::getInt32Ty(Mod->getContext())),
    TID
  };

  Value *Offset = Builder.CreateInBoundsGEP(GVTy, GV, Indices);
  I.mutateType(Offset->getType());
  I.replaceAllUsesWith(Offset);
  I.eraseFromParent();

  for (Value *V : WorkList) {
    CallInst *Call = dyn_cast<CallInst>(V);
    if (!Call) {
      if (ICmpInst *CI = dyn_cast<ICmpInst>(V)) {
        Value *Src0 = CI->getOperand(0);
        Type *EltTy = Src0->getType()->getPointerElementType();
        PointerType *NewTy = PointerType::get(EltTy, AS.LOCAL_ADDRESS);

        if (isa<ConstantPointerNull>(CI->getOperand(0)))
          CI->setOperand(0, ConstantPointerNull::get(NewTy));

        if (isa<ConstantPointerNull>(CI->getOperand(1)))
          CI->setOperand(1, ConstantPointerNull::get(NewTy));

        continue;
      }

      // The operand's value should be corrected on its own and we don't want to
      // touch the users.
      if (isa<AddrSpaceCastInst>(V))
        continue;

      Type *EltTy = V->getType()->getPointerElementType();
      PointerType *NewTy = PointerType::get(EltTy, AS.LOCAL_ADDRESS);

      // FIXME: It doesn't really make sense to try to do this for all
      // instructions.
      V->mutateType(NewTy);

      // Adjust the types of any constant operands.

//.........这里部分代码省略.........
    if (!URoR) return CleanupSelectors(CatchAllSels);
  }

  SmallPtrSet<InvokeInst*, 32> URoRInvokes;
  FindAllURoRInvokes(URoRInvokes);

  SmallPtrSet<IntrinsicInst*, 32> SelsToConvert;

  for (SmallPtrSet<IntrinsicInst*, 32>::iterator
         SI = Sels.begin(), SE = Sels.end(); SI != SE; ++SI) {
    const BasicBlock *SelBB = (*SI)->getParent();
    for (SmallPtrSet<InvokeInst*, 32>::iterator
           UI = URoRInvokes.begin(), UE = URoRInvokes.end(); UI != UE; ++UI) {
      const BasicBlock *URoRBB = (*UI)->getParent();
      if (DT->dominates(SelBB, URoRBB)) {
        SelsToConvert.insert(*SI);
        break;
      }
    }
  }

  bool Changed = false;

  if (Sels.size() != SelsToConvert.size()) {
    // If we haven't been able to convert all of the clean-up selectors, then
    // loop through the slow way to see if they still need to be converted.
    if (!ExceptionValueIntrinsic) {
      ExceptionValueIntrinsic =
        Intrinsic::getDeclaration(F->getParent(), Intrinsic::eh_exception);
      if (!ExceptionValueIntrinsic)
        return CleanupSelectors(CatchAllSels);
    }

    for (Value::use_iterator
           I = ExceptionValueIntrinsic->use_begin(),
           E = ExceptionValueIntrinsic->use_end(); I != E; ++I) {
      IntrinsicInst *EHPtr = dyn_cast<IntrinsicInst>(*I);
      if (!EHPtr || EHPtr->getParent()->getParent() != F) continue;

      Changed |= PromoteEHPtrStore(EHPtr);

      bool URoRInvoke = false;
      SmallPtrSet<IntrinsicInst*, 8> SelCalls;
      Changed |= FindSelectorAndURoR(EHPtr, URoRInvoke, SelCalls);

      if (URoRInvoke) {
        // This EH pointer is being used by an invoke of an URoR instruction and
        // an eh.selector intrinsic call. If the eh.selector is a 'clean-up', we
        // need to convert it to a 'catch-all'.
        for (SmallPtrSet<IntrinsicInst*, 8>::iterator
               SI = SelCalls.begin(), SE = SelCalls.end(); SI != SE; ++SI)
          if (!HasCatchAllInSelector(*SI))
              SelsToConvert.insert(*SI);
      }
    }
  }

  if (!SelsToConvert.empty()) {
    // Convert all clean-up eh.selectors, which are associated with "invokes" of
    // URoR calls, into catch-all eh.selectors.
    Changed = true;

    for (SmallPtrSet<IntrinsicInst*, 8>::iterator
           SI = SelsToConvert.begin(), SE = SelsToConvert.end();
         SI != SE; ++SI) {
      IntrinsicInst *II = *SI;

      // Use the exception object pointer and the personality function
      // from the original selector.
      CallSite CS(II);
      IntrinsicInst::op_iterator I = CS.arg_begin();
      IntrinsicInst::op_iterator E = CS.arg_end();
      IntrinsicInst::op_iterator B = prior(E);

      // Exclude last argument if it is an integer.
      if (isa<ConstantInt>(B)) E = B;

      // Add exception object pointer (front).
      // Add personality function (next).
      // Add in any filter IDs (rest).
      SmallVector<Value*, 8> Args(I, E);

      Args.push_back(EHCatchAllValue->getInitializer()); // Catch-all indicator.

      CallInst *NewSelector =
        CallInst::Create(SelectorIntrinsic, Args.begin(), Args.end(),
                         "eh.sel.catch.all", II);

      NewSelector->setTailCall(II->isTailCall());
      NewSelector->setAttributes(II->getAttributes());
      NewSelector->setCallingConv(II->getCallingConv());

      II->replaceAllUsesWith(NewSelector);
      II->eraseFromParent();
    }
  }

  Changed |= CleanupSelectors(CatchAllSels);
  return Changed;
}

//.........这里部分代码省略.........
          Value *int_max_s =
            ConstantInt::get(op1->getType(), APInt::getSignedMaxValue(bw));
          Value *int_min_s =
            ConstantInt::get(op1->getType(), APInt::getSignedMinValue(bw));
          Value *int_max =
            builder.CreateSExt(int_max_s, IntegerType::get(M.getContext(), bw2));
          Value *int_min =
            builder.CreateSExt(int_min_s, IntegerType::get(M.getContext(), bw2));

          if (ii->getIntrinsicID() == Intrinsic::sadd_with_overflow){
            result_ext = builder.CreateAdd(op1ext, op2ext);
          } else if (ii->getIntrinsicID() == Intrinsic::ssub_with_overflow){
            result_ext = builder.CreateSub(op1ext, op2ext);
          } else if (ii->getIntrinsicID() == Intrinsic::smul_with_overflow){
            result_ext = builder.CreateMul(op1ext, op2ext);
          }
          overflow = builder.CreateOr(builder.CreateICmpSGT(result_ext, int_max),
                                      builder.CreateICmpSLT(result_ext, int_min));
        }

        // This trunc cound be replaced by a more general trunc replacement
        // that allows to detect also undefined behavior in assignments or
        // overflow in operation with integers whose dimension is smaller than
        // int's dimension, e.g.
        //     uint8_t = uint8_t + uint8_t;
        // if one desires the wrapping should write
        //     uint8_t = (uint8_t + uint8_t) & 0xFF;
        // before this, must check if it has side effects on other operations
        result = builder.CreateTrunc(result_ext, op1->getType());
        Value *resultStruct =
          builder.CreateInsertValue(UndefValue::get(ii->getType()), result, 0);
        resultStruct = builder.CreateInsertValue(resultStruct, overflow, 1);
        
        ii->replaceAllUsesWith(resultStruct);
        ii->removeFromParent();
        delete ii;
        dirty = true;
        break;
      }

      case Intrinsic::dbg_value:
      case Intrinsic::dbg_declare:
        // Remove these regardless of lower intrinsics flag. This can
        // be removed once IntrinsicLowering is fixed to not have bad
        // caches.
        ii->eraseFromParent();
        dirty = true;
        break;

      case Intrinsic::trap: {
        // Intrisic instruction "llvm.trap" found. Directly lower it to
        // a call of the abort() function.
        Function *F = cast<Function>(
          M.getOrInsertFunction(
            "abort", Type::getVoidTy(getGlobalContext()), NULL));
        F->setDoesNotReturn();
        F->setDoesNotThrow();

        CallInst::Create(F, Twine(), ii);
        new UnreachableInst(getGlobalContext(), ii);

        ii->eraseFromParent();

        dirty = true;
        break;
      }