C++ LoadInst::setMetadata方法代码示例

void Closure::unpack_struct(Scope<Value *> &dst,
                            llvm::Type *
#if LLVM_VERSION >= 37
                            type
#endif
                            ,
                            Value *src,
                            IRBuilder<> *builder) {
    // type, type of src should be a pointer to a struct of the type returned by build_type
    int idx = 0;
    LLVMContext &context = builder->getContext();
    vector<string> nm = names();
    for (size_t i = 0; i < nm.size(); i++) {
#if LLVM_VERSION >= 37
        Value *ptr = builder->CreateConstInBoundsGEP2_32(type, src, 0, idx++);
#else
        Value *ptr = builder->CreateConstInBoundsGEP2_32(src, 0, idx++);
#endif
        LoadInst *load = builder->CreateLoad(ptr);
        if (load->getType()->isPointerTy()) {
            // Give it a unique type so that tbaa tells llvm that this can't alias anything
            LLVMMDNodeArgumentType md_args[] = {MDString::get(context, nm[i])};
            load->setMetadata("tbaa", MDNode::get(context, md_args));
        }
        dst.push(nm[i], load);
        load->setName(nm[i]);
    }
}

bool AMDGPUCodeGenPrepare::visitLoadInst(LoadInst &I) {
  if (!WidenLoads)
    return false;

  if ((I.getPointerAddressSpace() == AMDGPUASI.CONSTANT_ADDRESS ||
       I.getPointerAddressSpace() == AMDGPUASI.CONSTANT_ADDRESS_32BIT) &&
      canWidenScalarExtLoad(I)) {
    IRBuilder<> Builder(&I);
    Builder.SetCurrentDebugLocation(I.getDebugLoc());

    Type *I32Ty = Builder.getInt32Ty();
    Type *PT = PointerType::get(I32Ty, I.getPointerAddressSpace());
    Value *BitCast= Builder.CreateBitCast(I.getPointerOperand(), PT);
    LoadInst *WidenLoad = Builder.CreateLoad(BitCast);
    WidenLoad->copyMetadata(I);

    // If we have range metadata, we need to convert the type, and not make
    // assumptions about the high bits.
    if (auto *Range = WidenLoad->getMetadata(LLVMContext::MD_range)) {
      ConstantInt *Lower =
        mdconst::extract<ConstantInt>(Range->getOperand(0));

      if (Lower->getValue().isNullValue()) {
        WidenLoad->setMetadata(LLVMContext::MD_range, nullptr);
      } else {
        Metadata *LowAndHigh[] = {
          ConstantAsMetadata::get(ConstantInt::get(I32Ty, Lower->getValue().zext(32))),
          // Don't make assumptions about the high bits.
          ConstantAsMetadata::get(ConstantInt::get(I32Ty, 0))
        };

        WidenLoad->setMetadata(LLVMContext::MD_range,
                               MDNode::get(Mod->getContext(), LowAndHigh));
      }
    }

    int TySize = Mod->getDataLayout().getTypeSizeInBits(I.getType());
    Type *IntNTy = Builder.getIntNTy(TySize);
    Value *ValTrunc = Builder.CreateTrunc(WidenLoad, IntNTy);
    Value *ValOrig = Builder.CreateBitCast(ValTrunc, I.getType());
    I.replaceAllUsesWith(ValOrig);
    I.eraseFromParent();
    return true;
  }

  return false;
}

void SanitizerCoverageModule::InjectCoverageAtBlock(Function &F,
                                                    BasicBlock &BB) {
  BasicBlock::iterator IP = BB.getFirstInsertionPt(), BE = BB.end();
  // Skip static allocas at the top of the entry block so they don't become
  // dynamic when we split the block.  If we used our optimized stack layout,
  // then there will only be one alloca and it will come first.
  for (; IP != BE; ++IP) {
    AllocaInst *AI = dyn_cast<AllocaInst>(IP);
    if (!AI || !AI->isStaticAlloca())
      break;
  }

  bool IsEntryBB = &BB == &F.getEntryBlock();
  DebugLoc EntryLoc =
      IsEntryBB ? IP->getDebugLoc().getFnDebugLoc(*C) : IP->getDebugLoc();
  IRBuilder<> IRB(IP);
  IRB.SetCurrentDebugLocation(EntryLoc);
  SmallVector<Value *, 1> Indices;
  Value *GuardP = IRB.CreateAdd(
      IRB.CreatePointerCast(GuardArray, IntptrTy),
      ConstantInt::get(IntptrTy, (1 + SanCovFunction->getNumUses()) * 4));
  Type *Int32PtrTy = PointerType::getUnqual(IRB.getInt32Ty());
  GuardP = IRB.CreateIntToPtr(GuardP, Int32PtrTy);
  LoadInst *Load = IRB.CreateLoad(GuardP);
  Load->setAtomic(Monotonic);
  Load->setAlignment(4);
  Load->setMetadata(F.getParent()->getMDKindID("nosanitize"),
                    MDNode::get(*C, None));
  Value *Cmp = IRB.CreateICmpSGE(Constant::getNullValue(Load->getType()), Load);
  Instruction *Ins = SplitBlockAndInsertIfThen(
      Cmp, IP, false, MDBuilder(*C).createBranchWeights(1, 100000));
  IRB.SetInsertPoint(Ins);
  IRB.SetCurrentDebugLocation(EntryLoc);
  // __sanitizer_cov gets the PC of the instruction using GET_CALLER_PC.
  IRB.CreateCall(SanCovFunction, GuardP);
  IRB.CreateCall(EmptyAsm);  // Avoids callback merge.

  if (ClExperimentalTracing) {
    // Experimental support for tracing.
    // Insert a callback with the same guard variable as used for coverage.
    IRB.SetInsertPoint(IP);
    IRB.CreateCall(IsEntryBB ? SanCovTraceEnter : SanCovTraceBB, GuardP);
  }
}

/// \brief Helper to combine a load to a new type.
///
/// This just does the work of combining a load to a new type. It handles
/// metadata, etc., and returns the new instruction. The \c NewTy should be the
/// loaded *value* type. This will convert it to a pointer, cast the operand to
/// that pointer type, load it, etc.
///
/// Note that this will create all of the instructions with whatever insert
/// point the \c InstCombiner currently is using.
static LoadInst *combineLoadToNewType(InstCombiner &IC, LoadInst &LI, Type *NewTy) {
  Value *Ptr = LI.getPointerOperand();
  unsigned AS = LI.getPointerAddressSpace();
  SmallVector<std::pair<unsigned, MDNode *>, 8> MD;
  LI.getAllMetadata(MD);

  LoadInst *NewLoad = IC.Builder->CreateAlignedLoad(
      IC.Builder->CreateBitCast(Ptr, NewTy->getPointerTo(AS)),
      LI.getAlignment(), LI.getName());
  for (const auto &MDPair : MD) {
    unsigned ID = MDPair.first;
    MDNode *N = MDPair.second;
    // Note, essentially every kind of metadata should be preserved here! This
    // routine is supposed to clone a load instruction changing *only its type*.
    // The only metadata it makes sense to drop is metadata which is invalidated
    // when the pointer type changes. This should essentially never be the case
    // in LLVM, but we explicitly switch over only known metadata to be
    // conservatively correct. If you are adding metadata to LLVM which pertains
    // to loads, you almost certainly want to add it here.
    switch (ID) {
    case LLVMContext::MD_dbg:
    case LLVMContext::MD_tbaa:
    case LLVMContext::MD_prof:
    case LLVMContext::MD_fpmath:
    case LLVMContext::MD_tbaa_struct:
    case LLVMContext::MD_invariant_load:
    case LLVMContext::MD_alias_scope:
    case LLVMContext::MD_noalias:
    case LLVMContext::MD_nontemporal:
    case LLVMContext::MD_mem_parallel_loop_access:
    case LLVMContext::MD_nonnull:
      // All of these directly apply.
      NewLoad->setMetadata(ID, N);
      break;

    case LLVMContext::MD_range:
      // FIXME: It would be nice to propagate this in some way, but the type
      // conversions make it hard.
      break;
    }
  }
  return NewLoad;
}

void Closure::unpack_struct(Scope<Value *> &dst,
                            Value *src,
                            IRBuilder<> *builder) {
    // src should be a pointer to a struct of the type returned by build_type
    int idx = 0;
    LLVMContext &context = builder->getContext();
    vector<string> nm = names();
    for (size_t i = 0; i < nm.size(); i++) {
        Value *ptr = builder->CreateConstInBoundsGEP2_32(src, 0, idx++);
        LoadInst *load = builder->CreateLoad(ptr);
        if (load->getType()->isPointerTy()) {
            // Give it a unique type so that tbaa tells llvm that this can't alias anything
            load->setMetadata("tbaa", MDNode::get(context,
                                                  vec<Value *>(MDString::get(context, nm[i]))));
        }
        dst.push(nm[i], load);
        load->setName(nm[i]);
    }
}

std::pair<Value *, Value *>
AMDGPUPromoteAlloca::getLocalSizeYZ(IRBuilder<> &Builder) {
  if (!IsAMDHSA) {
    Function *LocalSizeYFn
      = Intrinsic::getDeclaration(Mod, Intrinsic::r600_read_local_size_y);
    Function *LocalSizeZFn
      = Intrinsic::getDeclaration(Mod, Intrinsic::r600_read_local_size_z);

    CallInst *LocalSizeY = Builder.CreateCall(LocalSizeYFn, {});
    CallInst *LocalSizeZ = Builder.CreateCall(LocalSizeZFn, {});

    LocalSizeY->setMetadata(LLVMContext::MD_range, MaxWorkGroupSizeRange);
    LocalSizeZ->setMetadata(LLVMContext::MD_range, MaxWorkGroupSizeRange);

    return std::make_pair(LocalSizeY, LocalSizeZ);
  }

  // We must read the size out of the dispatch pointer.
  assert(IsAMDGCN);

  // We are indexing into this struct, and want to extract the workgroup_size_*
  // fields.
  //
  //   typedef struct hsa_kernel_dispatch_packet_s {
  //     uint16_t header;
  //     uint16_t setup;
  //     uint16_t workgroup_size_x ;
  //     uint16_t workgroup_size_y;
  //     uint16_t workgroup_size_z;
  //     uint16_t reserved0;
  //     uint32_t grid_size_x ;
  //     uint32_t grid_size_y ;
  //     uint32_t grid_size_z;
  //
  //     uint32_t private_segment_size;
  //     uint32_t group_segment_size;
  //     uint64_t kernel_object;
  //
  // #ifdef HSA_LARGE_MODEL
  //     void *kernarg_address;
  // #elif defined HSA_LITTLE_ENDIAN
  //     void *kernarg_address;
  //     uint32_t reserved1;
  // #else
  //     uint32_t reserved1;
  //     void *kernarg_address;
  // #endif
  //     uint64_t reserved2;
  //     hsa_signal_t completion_signal; // uint64_t wrapper
  //   } hsa_kernel_dispatch_packet_t
  //
  Function *DispatchPtrFn
    = Intrinsic::getDeclaration(Mod, Intrinsic::amdgcn_dispatch_ptr);

  CallInst *DispatchPtr = Builder.CreateCall(DispatchPtrFn, {});
  DispatchPtr->addAttribute(AttributeSet::ReturnIndex, Attribute::NoAlias);
  DispatchPtr->addAttribute(AttributeSet::ReturnIndex, Attribute::NonNull);

  // Size of the dispatch packet struct.
  DispatchPtr->addDereferenceableAttr(AttributeSet::ReturnIndex, 64);

  Type *I32Ty = Type::getInt32Ty(Mod->getContext());
  Value *CastDispatchPtr = Builder.CreateBitCast(
    DispatchPtr, PointerType::get(I32Ty, AMDGPUAS::CONSTANT_ADDRESS));

  // We could do a single 64-bit load here, but it's likely that the basic
  // 32-bit and extract sequence is already present, and it is probably easier
  // to CSE this. The loads should be mergable later anyway.
  Value *GEPXY = Builder.CreateConstInBoundsGEP1_64(CastDispatchPtr, 1);
  LoadInst *LoadXY = Builder.CreateAlignedLoad(GEPXY, 4);

  Value *GEPZU = Builder.CreateConstInBoundsGEP1_64(CastDispatchPtr, 2);
  LoadInst *LoadZU = Builder.CreateAlignedLoad(GEPZU, 4);

  MDNode *MD = llvm::MDNode::get(Mod->getContext(), None);
  LoadXY->setMetadata(LLVMContext::MD_invariant_load, MD);
  LoadZU->setMetadata(LLVMContext::MD_invariant_load, MD);
  LoadZU->setMetadata(LLVMContext::MD_range, MaxWorkGroupSizeRange);

  // Extract y component. Upper half of LoadZU should be zero already.
  Value *Y = Builder.CreateLShr(LoadXY, 16);

  return std::make_pair(Y, LoadZU);
}

/// \brief Helper to combine a load to a new type.
///
/// This just does the work of combining a load to a new type. It handles
/// metadata, etc., and returns the new instruction. The \c NewTy should be the
/// loaded *value* type. This will convert it to a pointer, cast the operand to
/// that pointer type, load it, etc.
///
/// Note that this will create all of the instructions with whatever insert
/// point the \c InstCombiner currently is using.
static LoadInst *combineLoadToNewType(InstCombiner &IC, LoadInst &LI, Type *NewTy,
                                      const Twine &Suffix = "") {
  Value *Ptr = LI.getPointerOperand();
  unsigned AS = LI.getPointerAddressSpace();
  SmallVector<std::pair<unsigned, MDNode *>, 8> MD;
  LI.getAllMetadata(MD);

  LoadInst *NewLoad = IC.Builder->CreateAlignedLoad(
      IC.Builder->CreateBitCast(Ptr, NewTy->getPointerTo(AS)),
      LI.getAlignment(), LI.getName() + Suffix);
  MDBuilder MDB(NewLoad->getContext());
  for (const auto &MDPair : MD) {
    unsigned ID = MDPair.first;
    MDNode *N = MDPair.second;
    // Note, essentially every kind of metadata should be preserved here! This
    // routine is supposed to clone a load instruction changing *only its type*.
    // The only metadata it makes sense to drop is metadata which is invalidated
    // when the pointer type changes. This should essentially never be the case
    // in LLVM, but we explicitly switch over only known metadata to be
    // conservatively correct. If you are adding metadata to LLVM which pertains
    // to loads, you almost certainly want to add it here.
    switch (ID) {
    case LLVMContext::MD_dbg:
    case LLVMContext::MD_tbaa:
    case LLVMContext::MD_prof:
    case LLVMContext::MD_fpmath:
    case LLVMContext::MD_tbaa_struct:
    case LLVMContext::MD_invariant_load:
    case LLVMContext::MD_alias_scope:
    case LLVMContext::MD_noalias:
    case LLVMContext::MD_nontemporal:
    case LLVMContext::MD_mem_parallel_loop_access:
      // All of these directly apply.
      NewLoad->setMetadata(ID, N);
      break;

    case LLVMContext::MD_nonnull:
      // This only directly applies if the new type is also a pointer.
      if (NewTy->isPointerTy()) {
        NewLoad->setMetadata(ID, N);
        break;
      }
      // If it's integral now, translate it to !range metadata.
      if (NewTy->isIntegerTy()) {
        auto *ITy = cast<IntegerType>(NewTy);
        auto *NullInt = ConstantExpr::getPtrToInt(
            ConstantPointerNull::get(cast<PointerType>(Ptr->getType())), ITy);
        auto *NonNullInt =
            ConstantExpr::getAdd(NullInt, ConstantInt::get(ITy, 1));
        NewLoad->setMetadata(LLVMContext::MD_range,
                             MDB.createRange(NonNullInt, NullInt));
      }
      break;
    case LLVMContext::MD_align:
    case LLVMContext::MD_dereferenceable:
    case LLVMContext::MD_dereferenceable_or_null:
      // These only directly apply if the new type is also a pointer.
      if (NewTy->isPointerTy())
        NewLoad->setMetadata(ID, N);
      break;
    case LLVMContext::MD_range:
      // FIXME: It would be nice to propagate this in some way, but the type
      // conversions make it hard. If the new type is a pointer, we could
      // translate it to !nonnull metadata.
      break;
    }
  }
  return NewLoad;
}

//.........这里部分代码省略.........
        ScalarizeTable &ArgIndices = ScalarizedElements[I];
        // Store the Value* version of the indices in here, but declare it now
        // for reuse.
        std::vector<Value*> Ops;
        for (ScalarizeTable::iterator SI = ArgIndices.begin(),
               E = ArgIndices.end(); SI != E; ++SI) {
          Value *V = *AI;
          LoadInst *OrigLoad = OriginalLoads[*SI];
          if (!SI->empty()) {
            Ops.reserve(SI->size());
            const Type *ElTy = V->getType();
            for (IndicesVector::const_iterator II = SI->begin(),
                 IE = SI->end(); II != IE; ++II) {
              // Use i32 to index structs, and i64 for others (pointers/arrays).
              // This satisfies GEP constraints.
              const Type *IdxTy = (ElTy->isStructTy() ?
                    Type::getInt32Ty(F->getContext()) : 
                    Type::getInt64Ty(F->getContext()));
              Ops.push_back(ConstantInt::get(IdxTy, *II));
              // Keep track of the type we're currently indexing.
              ElTy = cast<CompositeType>(ElTy)->getTypeAtIndex(*II);
            }
            // And create a GEP to extract those indices.
            V = GetElementPtrInst::Create(V, Ops.begin(), Ops.end(),
                                          V->getName()+".idx", Call);
            Ops.clear();
            AA.copyValue(OrigLoad->getOperand(0), V);
          }
          // Since we're replacing a load make sure we take the alignment
          // of the previous load.
          LoadInst *newLoad = new LoadInst(V, V->getName()+".val", Call);
          newLoad->setAlignment(OrigLoad->getAlignment());
          // Transfer the TBAA info too.
          newLoad->setMetadata(LLVMContext::MD_tbaa,
                               OrigLoad->getMetadata(LLVMContext::MD_tbaa));
          Args.push_back(newLoad);
          AA.copyValue(OrigLoad, Args.back());
        }
      }

    if (ExtraArgHack)
      Args.push_back(Constant::getNullValue(Type::getInt32Ty(F->getContext())));

    // Push any varargs arguments on the list.
    for (; AI != CS.arg_end(); ++AI, ++ArgIndex) {
      Args.push_back(*AI);
      if (Attributes Attrs = CallPAL.getParamAttributes(ArgIndex))
        AttributesVec.push_back(AttributeWithIndex::get(Args.size(), Attrs));
    }

    // Add any function attributes.
    if (Attributes attrs = CallPAL.getFnAttributes())
      AttributesVec.push_back(AttributeWithIndex::get(~0, attrs));

    Instruction *New;
    if (InvokeInst *II = dyn_cast<InvokeInst>(Call)) {
      New = InvokeInst::Create(NF, II->getNormalDest(), II->getUnwindDest(),
                               Args.begin(), Args.end(), "", Call);
      cast<InvokeInst>(New)->setCallingConv(CS.getCallingConv());
      cast<InvokeInst>(New)->setAttributes(AttrListPtr::get(AttributesVec.begin(),
                                                          AttributesVec.end()));
    } else {
      New = CallInst::Create(NF, Args.begin(), Args.end(), "", Call);
      cast<CallInst>(New)->setCallingConv(CS.getCallingConv());
      cast<CallInst>(New)->setAttributes(AttrListPtr::get(AttributesVec.begin(),
                                                        AttributesVec.end()));

//.........这里部分代码省略.........

    int64_t AlignDownOffset = alignDown(EltOffset, 4);
    int64_t OffsetDiff = EltOffset - AlignDownOffset;
    unsigned AdjustedAlign = MinAlign(DoShiftOpt ? AlignDownOffset : EltOffset,
                                      KernArgBaseAlign);

    Value *ArgPtr;
    Type *AdjustedArgTy;
    if (DoShiftOpt) { // FIXME: Handle aggregate types
      // Since we don't have sub-dword scalar loads, avoid doing an extload by
      // loading earlier than the argument address, and extracting the relevant
      // bits.
      //
      // Additionally widen any sub-dword load to i32 even if suitably aligned,
      // so that CSE between different argument loads works easily.
      ArgPtr = Builder.CreateConstInBoundsGEP1_64(
          Builder.getInt8Ty(), KernArgSegment, AlignDownOffset,
          Arg.getName() + ".kernarg.offset.align.down");
      AdjustedArgTy = Builder.getInt32Ty();
    } else {
      ArgPtr = Builder.CreateConstInBoundsGEP1_64(
          Builder.getInt8Ty(), KernArgSegment, EltOffset,
          Arg.getName() + ".kernarg.offset");
      AdjustedArgTy = ArgTy;
    }

    if (IsV3 && Size >= 32) {
      V4Ty = VectorType::get(VT->getVectorElementType(), 4);
      // Use the hack that clang uses to avoid SelectionDAG ruining v3 loads
      AdjustedArgTy = V4Ty;
    }

    ArgPtr = Builder.CreateBitCast(ArgPtr, AdjustedArgTy->getPointerTo(AS),
                                   ArgPtr->getName() + ".cast");
    LoadInst *Load =
        Builder.CreateAlignedLoad(AdjustedArgTy, ArgPtr, AdjustedAlign);
    Load->setMetadata(LLVMContext::MD_invariant_load, MDNode::get(Ctx, {}));

    MDBuilder MDB(Ctx);

    if (isa<PointerType>(ArgTy)) {
      if (Arg.hasNonNullAttr())
        Load->setMetadata(LLVMContext::MD_nonnull, MDNode::get(Ctx, {}));

      uint64_t DerefBytes = Arg.getDereferenceableBytes();
      if (DerefBytes != 0) {
        Load->setMetadata(
          LLVMContext::MD_dereferenceable,
          MDNode::get(Ctx,
                      MDB.createConstant(
                        ConstantInt::get(Builder.getInt64Ty(), DerefBytes))));
      }

      uint64_t DerefOrNullBytes = Arg.getDereferenceableOrNullBytes();
      if (DerefOrNullBytes != 0) {
        Load->setMetadata(
          LLVMContext::MD_dereferenceable_or_null,
          MDNode::get(Ctx,
                      MDB.createConstant(ConstantInt::get(Builder.getInt64Ty(),
                                                          DerefOrNullBytes))));
      }

      unsigned ParamAlign = Arg.getParamAlignment();
      if (ParamAlign != 0) {
        Load->setMetadata(
          LLVMContext::MD_align,
          MDNode::get(Ctx,
                      MDB.createConstant(ConstantInt::get(Builder.getInt64Ty(),
                                                          ParamAlign))));
      }
    }

    // TODO: Convert noalias arg to !noalias

    if (DoShiftOpt) {
      Value *ExtractBits = OffsetDiff == 0 ?
        Load : Builder.CreateLShr(Load, OffsetDiff * 8);

      IntegerType *ArgIntTy = Builder.getIntNTy(Size);
      Value *Trunc = Builder.CreateTrunc(ExtractBits, ArgIntTy);
      Value *NewVal = Builder.CreateBitCast(Trunc, ArgTy,
                                            Arg.getName() + ".load");
      Arg.replaceAllUsesWith(NewVal);
    } else if (IsV3) {
      Value *Shuf = Builder.CreateShuffleVector(Load, UndefValue::get(V4Ty),
                                                {0, 1, 2},
                                                Arg.getName() + ".load");
      Arg.replaceAllUsesWith(Shuf);
    } else {
      Load->setName(Arg.getName() + ".load");
      Arg.replaceAllUsesWith(Load);
    }
  }

  KernArgSegment->addAttribute(
    AttributeList::ReturnIndex,
    Attribute::getWithAlignment(Ctx, std::max(KernArgBaseAlign, MaxAlign)));

  return true;
}